Description

reference pytorch code for named entity tagging.

embedding
- word : GloVe, BERT, DistilBERT, mDistilBERT, MiniLM, feature-based BERT using DSA(Dynamic Self Attention) pooling, SpanBERT, ALBERT, RoBERTa, XLM-RoBERTa, BART, ELECTRA, DeBERTa, ELMo
- character : CNN
- pos : lookup
encoding
- BiLSTM
- DenseNet
- Dynamic Self-Attention: Computing Attention over Words Dynamically for Sentence Embedding
- a slightly modified DenseNet for longer dependency.
- Multi-Head Attention
decoding
- Softmax, CRF
related: reference pytorch code for intent(sentence) classification
document context for BERT paper reproduction
- see : https://github.com/dsindex/ntagger/issues/4#issuecomment-810304253
multi-task learning
- joint learning of sequence and token classification

Requirements

python >= 3.6
pip install -r requirements.txt

Data

CoNLL 2003 (English)

from etagger, CrossWeigh

data/conll2003

data/conll++

since CoNLL++/test.txt has incorrect chunk tags, combine it with original CoNLL2003/test.txt

$ python combine.py --conll2003 ../conll2003/test.txt --conllpp test.txt > t
$ mv t test.txt

data/conll2003_truecase, data/conll++_truecase

converting conll2003, conll++ data to its truecase
details

$ cd data/conll2003_truecase
$ python to-truecase.py --input_path ../conll2003/train.txt > train.txt
$ python to-truecase.py --input_path ../conll2003/valid.txt > valid.txt
$ python to-truecase.py --input_path ../conll2003/test.txt > test.txt

* same work for data/conll++

Kaggle NER (English)

from entity-annotated-corpus

data/kaggle

converting to CoNLL data format.
details

* download : https://www.kaggle.com/abhinavwalia95/entity-annotated-corpus?select=ner_dataset.csv
* remove illegal characters
$ sed -e 's/5 storm/storm/' ner_dataset.csv > t ; mv t ner_dataset.csv
$ iconv -f ISO-8859-1 -t UTF-8 ner_dataset.csv > ner_dataset.csv.utf

$ python to-conll.py
$ cp -rf valid.txt test.txt

GUM (English)

from entity-recognition-datasets

data/gum

converting to CoNLL data format.
details

* remove '*-object', '*-abstract'
$ python to-conll.py --input_train=gum-train.conll --inpu_test=gum-test.conll --train=train.txt --valid=valid.txt --test=test.txt

Naver NER 2019 (Korean)

from HanBert-NER

data/clova2019

converted to CoNLL data format.
details

이기범 eoj - B-PER
한두 eoj - O
쪽을 eoj - O
먹고 eoj - O
10분 eoj - B-TIM
후쯤 eoj - I-TIM
화제인을 eoj - B-CVL
먹는 eoj - O
것이 eoj - O
좋다고 eoj - O
한다 eoj - O
. eoj - O

data/clova2019_morph

tokenized by morphological analyzer and converted to CoNLL data format.
details

이기범 NNP - B-PER
한두 NNP - O
쪽 NNB - O
을 X-JKO - O
먹다 VV - O
고 X-EC - O
10 SN - B-TIM
분 X-NNB - I-TIM
후 NNG - I-TIM
쯤 X-XSN - I-TIM
화제 NNG - B-CVL
인 X-NNG - I-CVL
을 X-JKO - I-CVL
먹다 VV - O
...

'X-' prefix is prepending to POS(Part of Speech) tag of inside morphs for distinguishing following morphs.
we can evaluate the predicted result morph-by-morph or eojeol by eojeol manner(every lines having 'X-' POS tag are removed).

details

이기범 NNP - B-PER
한두 NNP - O
쪽 NNB - O
먹다 VV - O
10 SN - B-TIM
후 NNG - I-TIM
화제 NNG - B-CVL
먹다 VV - O
...

data/clova2019_morph_space

this data is identical to data/clova2019_morph except it treats spaces as tokens.
details

이기범 NNP - B-PER
_ _ - O
한두 NNP - O
_ _ - O
쪽 NNB - O
을 X-JKO - O
_ _ - O
먹다 VV - O
고 X-EC - O
_ _ - O
10 SN - B-TIM
분 X-NNB - I-TIM
_ _ - I-TIM
후 NNG - I-TIM
쯤 X-XSN - I-TIM
_ _ - O
화제 NNG - B-CVL
인 X-NNG - I-CVL
을 X-JKO - I-CVL
_ _ - O
먹다 VV - O
...

KMOU NER (Korean)

from KMOU NER

data/kmou2019

build train.raw, valid.raw
- data version : https://github.com/kmounlp/NER/commit/0b32e066870bda9f65cc190f5e89c2edc6cf8f6d
- same as pytorch-bert-crf-ner
- train.raw : 00002_NER.txt, ..., EXOBRAIN_NE_CORPUS_007.txt (1,425 files)
- valid.raw : EXOBRAIN_NE_CORPUS_009.txt, EXOBRAIN_NE_CORPUS_010.txt (2 files)
- apply correction rules and converting to CoNLL data format
  details

$ cd data/kmou2019
$ python correction.py -g train.raw > t
$ python to-conll.py -g t > train.txt

ex)
마음  마음  NNG B-POH
’   ’   SS  O
에   에   JKB O
_   _   _   O
담긴  담기+ㄴ    VV+ETM  O
->
마음 NNG - B-POH
’ SS - O
에 JKB - O
_ _ - O
담기다 VV - O
ㄴ ETM - O

$ python correction.py -g valid.raw > t
$ python to-conll.py -g t > valid.txt
$ cp -rf valid.txt test.txt

data/kmou2021

data source

https://github.com/kmounlp/NER/tree/master/%EA%B0%9C%EC%B2%B4%EB%AA%85%20%EC%9D%B8%EC%8B%9D/data/example/KMOU_BIO

testa.pos.txt  testa.tags.txt  testa.words.txt  testb.pos.txt  testb.tags.txt  testb.words.txt  train.pos.txt  train.tags.txt  train.words.txt

convert to CoNLL format

$ python to-conll.py --words=train.words.txt --tags=train.tags.txt --pos=train.pos.txt > train.txt
$ python to-conll.py --words=testa.words.txt --tags=testa.tags.txt --pos=testa.pos.txt > valid.txt
$ python to-conll.py --words=testb.words.txt --tags=testb.tags.txt --pos=testb.pos.txt > test.txt

KLUE NER (Korean)

from KLUE-benchmark

data/klue

convert to CoNLL format, original, character-based

$ python to-conll.py --file klue-ner-v1.1_train.tsv > train.txt
$ python to-conll.py --file klue-ner-v1.1_dev.tsv > valid.txt
$ python to-conll.py --file klue-ner-v1.1_dev.tsv > test.txt

convert to CoNLL format, segmented by khaiii

details

$ python to-conll.py --file klue-ner-v1.1_train.tsv --segmentation > train.txt
$ python to-conll.py --file klue-ner-v1.1_dev.tsv --segmentation > valid.txt
$ python to-conll.py --file klue-ner-v1.1_dev.tsv --segmentation > test.txt

특히 pos pt O
_ _ pt O
영동고속도 pos pt B-LC
로 pos pt I-LC
_ _ pt O
강릉 pos pt B-LC
_ _ pt O
방향 pos pt O
_ _ pt O
문막 pos pt B-LC
휴게소 pos pt I-LC
에서 pos pt O
_ _ pt O
만종분기점 pos pt B-LC
까지 pos pt O
_ _ pt O
5 pos pt B-QT
km pos pt I-QT
_ _ pt O
구간 pos pt O
에 pos pt O
는 pos pt O
_ _ pt O
승용차 pos pt O
_ _ pt O
전용 pos pt O
_ _ pt O
임시 pos pt O
_ _ pt O
갓길차로제 pos pt O
를 pos pt O
_ _ pt O
운영 pos pt O
하 pos pt O
기 pos pt O
로 pos pt O
_ _ pt O
했다. pos pt O
...

Pretrained models

English

glove

download GloVe6B and unzip to 'embeddings' dir

$ mkdir embeddings
$ ls embeddings
glove.6B.zip
$ unzip glove.6B.zip

BERT-like models(huggingface's transformers)
SpanBERT
pretrained SpanBERT models are compatible with huggingface's BERT modele except 'bert.pooler.dense.weight', 'bert.pooler.dense.bias'.

ELMo(allennlp)

$ cd embeddings
$ curl -OL https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway_5.5B/elmo_2x4096_512_2048cnn_2xhighway_5.5B_weights.hdf5
$ curl -OL https://s3-us-west-2.amazonaws.com/allennlp/models/elmo/2x4096_512_2048cnn_2xhighway_5.5B/elmo_2x4096_512_2048cnn_2xhighway_5.5B_options.json

Korean
- description of Korean GloVe, BERT, DistilBERT, ELECTRA
- GloVe : kor.glove.300k.300d.txt (inhouse)
- bpe BERT : kor-bert-base-bpe.v1, kor-bert-large-bpe.v1, v3 (inhouse)
- dha-bpe BERT : kor-bert-base-dha_bpe.v1, v3, kor-bert-large-dha_bpe.v1, v3 (inhouse)
- dha BERT : kor-bert-base-dha.v1, v2 (inhouse)
- MA-BERT-base : kor-bert-base-morpheme-aware (inhouse)
- KcBERT : kcbert-base, kcbert-large
- DistilBERT : kor-distil-bpe-bert.v1, kor-distil-dha-bert.v1, kor-distil-wp-bert.v1 (inhouse)
- mDistilBERT : distilbert-base-multilingual-cased
- KoELECTRA-Base : koelectra-base-v1-discriminator, koelectra-base-v3-discriminator
- LM-KOR-ELECTRA : electra-kor-base
- ELECTRA-base : kor-electra-bpe.v1, kor-electra-base-dhaToken1.large, kor-electra-base-dhaSyllable (inhouse)
- RoBERTa-base : kor-roberta-base-bbpe (inhouse)
- MA-RoBERTa-base : kor-roberta-base-morpheme-aware (inhouse)
- XLM-RoBERTa : xlm-roberta-base, xlm-roberta-large
- KLUE-RoBERTa : klue-roberta-base, klue-roberta-large
- Funnel-base : funnel-kor-base
- ELMo description
- kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_weights.hdf5, kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_options.json (inhouse)

CoNLL 2003 (English)

experiments summary

ntagger, measured by conlleval.pl (micro F1)

	F1 (%)	(truecase) F1 (%)	Features	GPU / CPU	ONNX	Dynamic	Etc
GloVe, BiLSTM	88.23		word, pos	5.6217 / -	3.6969		threads=14
GloVe, BiLSTM	88.18		word, pos	29.9140 / -	-		NCRF
GloVe, BiLSTM	88.94		word, character, pos	6.4108 / -	8.5656		threads=14
GloVe, BiLSTM-MHA	89.99		word, character, pos	7.7513 / -
GloVe, BiLSTM-MHA-CRF	90.48		word, character, pos	25.8200 / -
GloVe, BiLSTM-MHA-CRF	90.42		word, character, pos	30.1932 / -			LabelSmoothingCrossEntropy
GloVe, BiLSTM-CRF	90.14 / 90.92	90.26 / 90.76	word, character, pos	26.2807 / -	FAIL
ConceptNet, BiLSTM-CRF	87.78		word, character, pos	25.8119 / -
ConceptNet, BiLSTM-CRF	88.17		word, character, pos	23.3482 / -			optuna
GloVe, DenseNet-CRF	88.23		word, pos	24.7893 / -
GloVe, DenseNet-CRF	88.89		word, character, pos	28.0993 / -
GloVe, DenseNet-MHA	88.47		word, character, pos	7.9668 / -
GloVe, DenseNet-MHA-CRF	88.58		word, character, pos	26.2287 / -
BERT-tiny, BiLSTM	69.65		word	20.1376 / -
BERT-mini, BiLSTM	81.55		word	21.4632 / -
BERT-small, BiLSTM	86.35		word	22.6087 / -
BERT-medium, BiLSTM	88.29		word	27.0486 / -
DistilBERT, BiLSTM	89.50		word	13.4564 / -	56.2819	47.8320	threads=14
mDistilBERT, BiLSTM	90.21		word	13.4910 / -
MiniLM, BiLSTM	90.55		word	17.7890 / -
BERT-base(uncased), BiLSTM-CRF	90.20		word	42.6464 / -			epoch=10
BERT-base(uncased), BiLSTM	90.55		word	18.2323 / -	79.1914	83.9590	threads=14, epoch=10
BERT-base(uncased), BiLSTM	90.60		word	22.7757 / -			freezing BERT during some epochs, epoch=10
BERT-base(uncased), CRF	89.98		word	36.6893 / -			epoch=10
BERT-base(uncased)	90.25		word	16.6877 / -	72.8225	75.3025	threads=14, epoch=10
BERT-base(uncased), BiLSTM	89.03		word	24.9076 / -			del 8,9,10,11, threads=14, epoch=10
bert-base-NER(cased), BiLSTM	91.63 / 92.71	92.25 / 93.16	word	17.6680 / -			freezing BERT during some epochs for conll++,conll++_truecase, epoch=10
BERT-base(cased)	91.45		word	45.5024 / -			epoch=30
BERT-base(cased)	91.37		word	46.2147 / -			subword pooling, epoch=30
BERT-base(cased)	91.20		word	33.3276 / -			subword pooling, word embedding, epoch=30
BERT-base(cased)	92.35		word	32.6524 / -			document context, epoch=30, n_ctx=512
BERT-base(cased)	92.23		word	46.7166 / -			document context, subword pooling, epoch=30, n_ctx=512
BERT-base(cased)	92.15		word	46.5353 / -			document context, subword pooling, word embedding, epoch=30, n_ctx=512
BERT-base(cased), BiLSTM	90.20		word	21.5844 / -			epoch=10
BERT-base(cased), BiLSTM	90.99		word	21.7328 / -			freezing BERT during some epochs, epoch=10
BERT-base(cased), BiLSTM-MHA	90.95		word	21.9845 / -			freezing BERT during some epochs, epoch=10
BERT-base(cased), BiLSTM-CRF	90.17		word	43.4804 / -			epoch=10
BERT-base(cased), BiLSTM-CRF	91.55		word	42.2709 / -			freezing BERT during some epochs, epoch=10
BERT-base(cased), BiLSTM-CRF	91.60		word	39.6135 / -			using sub token label, freezing BERT during some epochs, epoch=10
BERT-base(cased), BiLSTM-CRF	91.33		word	41.1204 / -			slicing logits, freezing BERT during some epochs, epoch=10, https://github.com/dsindex/ntagger/releases/tag/v1.0
BERT-base(cased), BiLSTM-CRF	91.37		word, character, pos	40.2822 / -			using sub token label, freezing BERT during some epochs, epoch=10, https://github.com/dsindex/ntagger/releases/tag/v1.0
BERT-base(cased), BiLSTM-CRF	91.66		word, character, pos	39.6657 / -			using sub token label, freezing BERT during some epochs, epoch=30, https://github.com/dsindex/ntagger/releases/tag/v1.0
BERT-base(cased), BiLSTM-CRF	91.33		word	40.1493 / -			subword pooling, freezing BERT during some epochs, epoch=30
BERT-base(cased), BiLSTM-CRF	92.31		word	40.7793 / -			subword pooling, word embedding, freezing BERT during some epochs, epoch=30
BERT-base(cased), BiLSTM-CRF	92.08		word, character, pos	41.1466 / -			subword pooling, word embedding, freezing BERT during some epochs, epoch=30
BERT-base(cased), BiLSTM-CRF	92.04		word	92.8827 / -			document context, freezing BERT during some epochs, epoch=30, n_ctx=512
BERT-base(cased), BiLSTM-CRF	92.08		word	81.3581 / -			document context, subword pooling, freezing BERT during some epochs, epoch=30, n_ctx=512
BERT-base(cased), BiLSTM-CRF	92.85		word	80.1447 / -			document context, subword pooling, word embedding, freezing BERT during some epochs, epoch=30, n_ctx=512
BERT-large	91.13		word	31.2675 / -			epoch=30
BERT-large	92.27		word	68.4826 / -			document context, epoch=30, n_ctx=512
BERT-large, BiLSTM	91.32	91.89	word	40.3581 / -			epoch=10
BERT-large, BiLSTM	91.57		word	35.2808 / -			freezing BERT during some epochs, epoch=10
BERT-large, BiLSTM+CRF	90.78		word	59.3982 / -			epoch=10
BERT-large, BiLSTM+CRF	92.02	91.96	word	54.4254 / -			freezing BERT during some epochs, epoch=10
BERT-large, BiLSTM+CRF	92.83		word	103.930 / -			document context, subword pooling, word embedding, freezing BERT during some epochs, n_ctx=512, epoch=30
BERT-large, BiLSTM	89.10		word	33.1376 / -			del 12 ~ 23, epoch=10
BERT-large, BiLSTM-CRF	90.64		word	63.9397 / -			BERT as feature-based, [-4:] embedding, epoch=64
BERT-large, BiLSTM-CRF	90.52		word	70.8322 / -			BERT as feature-based, mean([0:3] + [-4:]) embedding, epoch=64
BERT-large, BiLSTM-CRF	90.81		word	68.6139 / -			BERT as feature-based, mean([0:17]) embedding, epoch=64
BERT-large, BiLSTM-CRF	90.76		word	60.8039 / -			BERT as feature-based, max([0:17]) embedding, epoch=64
BERT-large, BiLSTM-CRF	90.98		word	58.9112 / -			BERT as feature-based, mean([0:]) embedding, epoch=64
BERT-large, BiLSTM-CRF	90.62		word	66.6576 / -			BERT as feature-based, DSA(4, 300), epoch=64
BERT-large-squad, BiLSTM	91.75	92.17	word	35.6619 / -			epoch=10
BERT-large-conll03, BiLSTM	91.87 / 92.62	92.40 / 93.36	word	32.2211 / -			freezing BERT during some epochs for conll++,conll++_truecase, epoch=10
BERT-large-conll03	91.63	92.24	word	29.8476 / -			epoch=10
SpanBERT-base, BiLSTM	90.46		word	30.0991 / -
SpanBERT-large, BiLSTM	91.39	92.01	word	42.5959 / -
ALBERT-base, BiLSTM	88.19		word	31.0868 / -
ALBERT-xxlarge, BiLSTM	90.39		word	107.778 / -
RoBERTa-base	90.03		word	19.2503 / -			epoch=10
RoBERTa-large	91.83	91.90	word	28.5525 / -			epoch=10
XLM-RoBERTa-base	91.20		word	18.9604 / -			epoch=30
XLM-RoBERTa-base, BiLSTM	90.81		word	21.4667 / -			freezing BERT during some epochs, epoch=30
XLM-RoBERTa-base, BiLSTM-CRF	91.12		word	39.4418 / -			freezing BERT during some epochs, epoch=30
XLM-RoBERTa-base, BiLSTM-CRF	91.79		word	43.0662 / -			using sub token label, freezing BERT during some epochs, epoch=30
XLM-RoBERTa-base, BiLSTM-CRF	91.16		word	39.3642 / -			slicing logits, freezing BERT during some epochs, epoch=30, https://github.com/dsindex/ntagger/releases/tag/v1.0
XLM-RoBERTa-large	92.75 / 93.95	92.89 / 94.11	word	27.9144 / -			epoch=30
XLM-RoBERTa-large	93.86		word	50.2684 / -			document context, epoch=30, n_ctx=512
XLM-RoBERTa-large, BiLSTM	- / 93.75	- / 93.81	word	34.4894 / -			freezing BERT during some epochs, epoch=30
XLM-RoBERTa-large, BiLSTM-CRF	92.97		word	52.8133 / -			subword pooling, word embedding, freezing BERT during some epochs, epoch=30
XLM-RoBERTa-large, BiLSTM-CRF	93.59		word	105.188 / -			document context, subword pooling, word embedding, freezing BERT during some epochs, n_ctx=512, epoch=30
BART-large, BiLSTM	90.43		word	53.3657 / -
ELECTRA-base, BiLSTM	90.98		word	22.4132 / -
ELECTRA-large	91.39		word	29.5734 / -
DeBERTa-base	90.41		word	28.6874 / -			epoch=20
DeBERTa-large	91.45		word	53.9249 / -			epoch=20
DeBERTa-v2-xlarge	93.12		word	62.9722 / -			--use_sharded_ddp, epoch=20, https://github.com/dsindex/ntagger/releases/tag/v1.1
DeBERTa-v2-xlarge	94.00		word	157.075 / -			--use_sharded_ddp, document context, epoch=20, n_ctx=512, https://github.com/dsindex/ntagger/releases/tag/v1.1
ELMo, BiLSTM-CRF	91.78		word, pos	74.1001 / -
ELMo, BiLSTM-CRF	91.93		word, character, pos	67.6931 / -
ELMo, GloVe, BiLSTM-CRF	92.63 / 93.49	92.51 / 93.68	word, pos	74.6521 / -
ELMo, GloVe, BiLSTM-CRF	92.03		word, character, pos	60.4667 / -			threads=14

* GPU / CPU     : Elapsed time/example(ms), GPU / CPU, [Tesla V100 1 GPU, Intel(R) Xeon(R) Gold 5120 CPU @ 2.20GHz, 2 CPU, 14CORES/1CPU, HyperThreading]
* F1            : conll2003 / conll++
* (truecase) F1 : conll2003_truecase / conll++_truecase
* ONNX          : --enable_ort 
* Dynamic       : --enable_dqm
* default batch size, learning rate, n_ctx(max_seq_length) : 32, 1e-3, 180

etagger, measured by conlleval (micro F1)

	F1 (%)	(truecase) F1 (%)	Features	GPU / CPU	Etc
GloVe, BiLSTM-CRF	87.91		word	- / -
GloVe, BiLSTM-CRF	89.20		word, pos	14.9682 / 5.0336	LSTMBlockFusedCell(), threads=14
GloVe, BiLSTM-CRF	90.06		word, character, pos	15.8913 / 5.7952	LSTMBlockFusedCell(), threads=14
GloVe, BiLSTM-CRF	90.57		word, character, pos	24.6356 / 7.0887	LSTMCell(), threads=14
GloVe, BiLSTM-CRF	90.85		word, character, pos, chunk	- / -
GloVe, Transformer	88.62		word, character, pos, chunk	8.6126 / -	layers=4, heads=4, units=64
GloVe, Transformer-CRF	89.26		word, character, pos, chunk	16.528 / -	layers=4, heads=4, units=64
GloVe, Transformer-CRF	89.42		word, character, pos, chunk	19.157 / -	layers=6, heads=4, units=64
GloVe, Transformer-CRF	88.45		word, character, pos, chunk	30.534 / -	layers=6, heads=6, units=64
GloVe, Transformer-CRF	88.93		word, character, pos, chunk	20.257 / -	layers=6, heads=6, units=96
BERT-large, BiLSTM-CRF	90.22		word	- / -	BERT as feature-based
BERT-large, GloVe, BiLSTM-CRF	91.83		word	- / -	BERT as feature-based
ELMo, GloVe, BiLSTM-CRF	91.78		word, pos	- / -
ELMo, GloVe, BiLSTM-CRF	92.38		word, character, pos	46.4205 / 295.28	threads=14
ELMo, GloVe, BiLSTM-CRF	92.43		word, character, pos, chunk	- / -
ELMo, GloVe, BiLSTM-CRF	92.83	92.10	word, character, pos, chunk	- / -	GloVe-100d
BERT-large, ELMo, GloVe, BiLSTM-CRF	92.54		word, character, pos	- / -	BERT as feature-based, GloVe-100d

CoNLL 2003(English) leaderboard, measured by span-level F1 (micro F1, same result by conlleval? Unknown!)

	F1 (%)	Etc
LUKE	94.3
ACE + document-context	94.14
ACE	93.63
CNN Large + fine-tune	93.5
biaffine-ner	93.5
GCDT + BERT-L	93.47
LSTM-CRF+ELMo+BERT+Flair	93.38
Hierarchical + BERT	93.37
Flair embeddings + Pooling	93.18
BERT Large	92.8
BERT Base	92.4
BiLSTM-CRF+ELMo	92.22

CoNLL++(English) leaderboard, measured by span-level F1(micro F1, same result by conlleval? unknown!)

	F1 (%)	Etc
CrossWeigh + Pooled Flair	94.28
ELMo	93.42
BLSTM-CNN-CRF	91.87

emb_class=glove, enc_class=bilstm

- train ``` * token_emb_dim in configs/config-glove.json == 300 (ex, glove.6B.300d.txt ) $ python preprocess.py --data_dir=data/conll2003 * --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding $ python train.py --data_dir=data/conll2003 --use_crf * tensorboardX $ rm -rf runs $ tensorboard --logdir runs/ --port ${port} --bind_all ``` - evaluation ``` $ python evaluate.py --data_dir=data/conll2003 --use_crf * seqeval.metrics supports IOB2(BIO) format, so FB1 from conlleval.pl should be similar value with. $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8864423333037378, 3684 INFO:__main__:[Elapsed Time] : 97879ms, 26.54547922888949ms on average accuracy: 97.60%; precision: 88.90%; recall: 88.39%; FB1: 88.64 * --use_char_cnn INFO:__main__:[F1] : 0.9013611454834718, 3684 INFO:__main__:[Elapsed Time] : 96906ms, 26.280749389084985ms on average accuracy: 97.93%; precision: 89.99%; recall: 90.28%; FB1: 90.14 INFO:__main__:[F1] : 0.8981972428419936, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 95621.84238433838ms, 25.92995391408613ms on average accuracy: 97.82%; precision: 89.66%; recall: 89.98%; FB1: 89.82 * --data_dir=data/conll2003_truecase --use_char_cnn INFO:__main__:[F1] : 0.902609464838567, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 95198.03929328918ms, 25.81831880021024ms on average accuracy: 97.81%; precision: 90.19%; recall: 90.33%; FB1: 90.26 * --use_char_cnn --embedding_path=./embeddings/numberbatch-en-19.08.txt (from https://github.com/commonsense/conceptnet-numberbatch) INFO:__main__:[F1] : 0.877781702578594, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 95196.46453857422ms, 25.811972386352142ms on average accuracy: 97.47%; precision: 87.56%; recall: 88.00%; FB1: 87.78 * --use_char_cnn --lr=0.0025131520181464126 --batch_size=32 --seed=40 --embedding_path=./embeddings/numberbatch-en-19.08.txt (by optuna) INFO:__main__:[F1] : 0.8817204301075269, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 86105.42750358582ms, 23.34824522455005ms on average accuracy: 97.55%; precision: 87.79%; recall: 88.56%; FB1: 88.17 * --data_dir=data/conll++ --use_char_cnn INFO:__main__:[F1] : 0.9092189838865897, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 91787.76407241821ms, 24.89247232748379ms on average accuracy: 98.04%; precision: 91.30%; recall: 90.55%; FB1: 90.92 * --data_dir=data/conll++_truecase --use_char_cnn INFO:__main__:[F1] : 0.9076029567053854, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 91913.47312927246ms, 24.927138764347486ms on average accuracy: 98.01%; precision: 91.08%; recall: 90.44%; FB1: 90.76 * --use_char_cnn --use_mha , without --use_crf INFO:__main__:[F1] : 0.899894254494184, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 28661.02647781372ms, 7.751333820344053ms on average accuracy: 97.96%; precision: 89.58%; recall: 90.40%; FB1: 89.99 * --use_char_cnn --use_mha INFO:__main__:[F1] : 0.9047997878546804, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 95233.61730575562ms, 25.82004158968774ms on average accuracy: 97.97%; precision: 90.34%; recall: 90.62%; FB1: 90.48 * --use_char_cnn --use_mha --criterion=LabelSmoothingCrossEntropy INFO:__main__:[F1] : 0.9042186256301407, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 111375.56338310242ms, 30.19321803129985ms on average accuracy: 97.96%; precision: 90.33%; recall: 90.51%; FB1: 90.42 * --use_ncrf (use NCRF instead of pytorch-crf) $ python preprocess.py --data_dir=data/conll2003 --use_ncrf $ python train.py --data_dir=data/conll2003 --use_crf --use_ncrf $ python evaluate.py --data_dir=data/conll2003 --use_crf --use_ncrf INFO:__main__:[token classification F1] : 0.8818254461511142, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 110281.77309036255ms, 29.91403479502313ms on average accuracy: 97.40%; precision: 88.44%; recall: 87.92%; FB1: 88.18 $ python evaluate.py --data_dir=data/conll2003 --use_crf --use_ncrf --convert_onnx ... File "/usr/local/lib/python3.6/dist-packages/torch/onnx/symbolic_opset9.py", line 441, in transpose axes[dim0], axes[dim1] = axes[dim1], axes[dim0] IndexError: list index out of range ```

emb_class=glove, enc_class=densenet

- train ``` * token_emb_dim in configs/config-glove.json == 300 (ex, glove.6B.300d.txt ) $ python preprocess.py --config=configs/config-densenet.json --data_dir=data/conll2003 * --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding $ python train.py --config=configs/config-densenet.json --data_dir=data/conll2003 --save_path=pytorch-model-densenet.pt --use_crf --epoch=64 ``` - evaluation ``` $ python evaluate.py --config=configs/config-densenet.json --data_dir=data/conll2003 --model_path=pytorch-model-densenet.pt --use_crf $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8823112371499469, 3684 INFO:__main__:[Elapsed Time] : 91415ms, 24.789302199294053ms on average accuracy: 97.53%; precision: 88.33%; recall: 88.14%; FB1: 88.23 INFO:__main__:[F1] : 0.8736429969745506, 3684 INFO:__main__:[Elapsed Time] : 96335ms, 26.12136844963345ms on average accuracy: 97.36%; precision: 87.82%; recall: 86.92%; FB1: 87.36 * --use_char_cnn INFO:__main__:[F1] : 0.8847577092511013, 3684 INFO:__main__:[Elapsed Time] : 105399ms, 28.585120825414066ms on average accuracy: 97.60%; precision: 88.06%; recall: 88.90%; FB1: 88.48 INFO:__main__:[F1] : 0.8889086226800461, 3684 INFO:__main__:[Elapsed Time] : 103641ms, 28.099375509095847ms on average accuracy: 97.75%; precision: 89.17%; recall: 88.62%; FB1: 88.89 * --use_char_cnn --use_mha , without --use_crf INFO:__main__:[F1] : 0.8846794703148294, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 29449.145555496216ms, 7.9668397460655775ms on average accuracy: 97.75%; precision: 87.65%; recall: 89.31%; FB1: 88.47 * --use_char_cnn --use_mha INFO:__main__:[F1] : 0.8857545839210157, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 96725.40020942688ms, 26.228771176054657ms on average accuracy: 97.62%; precision: 88.20%; recall: 88.95%; FB1: 88.58 ```

emb_class=bert, enc_class=bilstm

- train ``` * n_ctx size should be less than 512 $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-large-cased * --use_crf for adding crf layer * --bert_use_pos for adding Part-Of-Speech features * --bert_use_feature_based for feature-based * --bert_disable_lstm for removing lstm layer $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-large-cased --bert_output_dir=bert-checkpoint --batch_size=16 --lr=1e-5 --epoch=10 ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9131544214694237, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 148789ms, 40.358131957643224ms on average accuracy: 98.27%; precision: 90.76%; recall: 91.87%; FB1: 91.32 * --batch_size=32 --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.9156838584756204, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 130123.88586997986ms, 35.28089417150219ms on average accuracy: 98.33%; precision: 91.04%; recall: 92.10%; FB1: 91.57 * --batch_size=32 INFO:__main__:[F1] : 0.9118733509234828, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 141879ms, 38.48330165625848ms on average accuracy: 98.24%; precision: 90.60%; recall: 91.78%; FB1: 91.19 * --bert_use_pos INFO:__main__:[F1] : 0.9111325554873234, 3684 INFO:__main__:[Elapsed Time] : 141093ms, 38.29885993485342ms on average accuracy: 98.24%; precision: 90.30%; recall: 91.94%; FB1: 91.11 * --use_crf INFO:__main__:[F1] : 0.9058430130235833, 3684 INFO:__main__:[Elapsed Time] : 218823ms, 59.398208469055376ms on average accuracy: 98.12%; precision: 90.44%; recall: 91.13%; FB1: 90.78 * --batch_size=32 --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --use_crf INFO:__main__:[F1] : 0.9178914019185074, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 194960.72125434875ms, 52.88945214113333ms on average accuracy: 98.31%; precision: 91.70%; recall: 92.33%; FB1: 92.02 * --data_dir=data/conll2003_truecase --batch_size=32 --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --use_crf INFO:__main__:[F1] : 0.9170551499692144, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 200677.77919769287ms, 54.42541078675742ms on average accuracy: 98.23%; precision: 91.63%; recall: 92.30%; FB1: 91.96 * --data_dir=data/conll2003_truecase INFO:__main__:[F1] : 0.9188571428571428, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 120905.13277053833ms, 32.785276922556875ms on average accuracy: 98.36%; precision: 91.25%; recall: 92.53%; FB1: 91.89 * --bert_disable_lstm --batch_size=32 INFO:__main__:[F1] : 0.9124989051414557, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 109073.38047027588ms, 29.574078060898998ms on average accuracy: 98.27%; precision: 90.29%; recall: 92.23%; FB1: 91.25 * --bert_disable_lstm --epoch=30 INFO:__main__:[F1] : 0.9112654999560285, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 115351.17220878601ms, 31.267560567345726ms on average accuracy: 98.30%; precision: 90.53%; recall: 91.73%; FB1: 91.13 * document context, --bert_disable_lstm --batch_size=8, n_ctx: 512, --epoch=30 # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-large-cased --bert_use_doc_context # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-large-cased --bert_output_dir=bert-checkpoint --batch_size=8 --lr=1e-5 --epoch=30 --bert_use_doc_context --bert_disable_lstm --eval_batch_size=32 # evaluate $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint --bert_use_doc_context --bert_disable_lstm $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9217056973086334, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 252364.78638648987ms, 68.47469475875883ms on average accuracy: 98.39%; precision: 90.99%; recall: 93.38%; FB1: 92.17 ** --bert_doc_context_option=2 # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-large-cased --bert_use_doc_context --bert_doc_context_option=2 INFO:__main__:[F1] : 0.9226704245199928, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 252395.40195465088ms, 68.4826407068746ms on average accuracy: 98.47%; precision: 91.80%; recall: 92.74%; FB1: 92.27 * document context, --bert_subword_pooling --bert_word_embedding --use_crf --batch_size=8, n_ctx: 512, --epoch=30 # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-large-cased --bert_use_doc_context --bert_use_subword_pooling --bert_use_word_embedding # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-large-cased --bert_output_dir=bert-checkpoint --batch_size=8 --lr=1e-5 --epoch=30 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_doc_context --bert_use_subword_pooling --bert_use_word_embedding --eval_batch_size=32 INFO:__main__:[F1] : 0.9283378987364141, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 382980.3364276886ms, 103.93010136099117ms on average accuracy: 98.48%; precision: 92.66%; recall: 93.01%; FB1: 92.83 * --bert_model_name_or_path=./embedings/bert-base-uncased --use_crf (BERT-base BiLSTM-CRF) INFO:__main__:[F1] : 0.8993429697766097, 3684 INFO:__main__:[Elapsed Time] : 100 examples, 4368ms, 42.64646464646464ms on average accuracy: 97.87%; precision: 89.52%; recall: 90.88%; FB1: 90.20 * --bert_model_name_or_path=./embedings/bert-base-uncased (BERT-base BiLSTM) INFO:__main__:[F1] : 0.9054532577903682, 3684 INFO:__main__:[Elapsed Time] : 100 examples, 1922ms, 18.232323232323232ms on average accuracy: 98.00%; precision: 90.55%; recall: 90.55%; FB1: 90.55 * --bert_model_name_or_path=./embedings/bert-base-uncased --warmup_epoch=0 --weight_decay=0.0 --epoch=20 (BERT-base BiLSTM) INFO:__main__:[F1] : 0.9049717912552891, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 68218ms, 18.403203909856096ms on average accuracy: 97.97%; precision: 90.12%; recall: 90.88%; FB1: 90.50 * --bert_model_name_or_path=./embedings/bert-base-uncased --batch_size=32 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.906032584764421, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 84019.82283592224ms, 22.77578745657356ms on average accuracy: 98.07%; precision: 90.13%; recall: 91.08%; FB1: 90.60 * --bert_model_name_or_path=./embedings/bert-base-uncased --batch_size=32 --epoch=20 --bert_freezing_epoch=7 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.9056670497745558, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 79642.83394813538ms, 21.5918773741234ms on average accuracy: 98.06%; precision: 90.45%; recall: 90.69%; FB1: 90.57 * --bert_model_name_or_path=./embedings/bert-base-uncased --bert_disable_lstm --use_crf (BERT-base CRF) INFO:__main__:[F1] : 0.8961356880573526, 3684 INFO:__main__:[Elapsed Time] : 135607ms, 36.68938365462938ms on average accuracy: 97.78%; precision: 89.24%; recall: 90.74%; FB1: 89.98 * --bert_model_name_or_path=./embedings/bert-base-uncased --bert_disable_lstm (BERT-base) INFO:__main__:[F1] : 0.9024668598015978, 3684 INFO:__main__:[Elapsed Time] : 61914ms, 16.68775454792289ms on average accuracy: 98.01%; precision: 89.50%; recall: 91.01%; FB1: 90.25 * https://huggingface.co/dslim/bert-base-NER * --bert_model_name_or_path=./embeddings/bert-base-NER --warmup_epoch=0 --weight_decay=0.0 --epoch=20 INFO:__main__:[F1] : 0.9163073132095397, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 65167.21606254578ms, 17.668077811708017ms on average accuracy: 98.34%; precision: 91.09%; recall: 92.17%; FB1: 91.63 * --data_dir=data/conll2003_truecase --bert_model_name_or_path=./embeddings/bert-base-NER --warmup_epoch=0 --weight_decay=0.0 --epoch=20 INFO:__main__:[F1] : 0.9224547212941797, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 66638.26131820679ms, 18.064501684967215ms on average accuracy: 98.38%; precision: 91.62%; recall: 92.88%; FB1: 92.25 * --data_dir=data/conll++ --bert_model_name_or_path=./embeddings/bert-base-NER --warmup_epoch=0 --weight_decay=0.0 --epoch=20 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.9271244409365955, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 82041.87273979187ms, 22.243568633776334ms on average accuracy: 98.53%; precision: 92.72%; recall: 92.70%; FB1: 92.71 * --data_dir=data/conll++_truecase --bert_model_name_or_path=./embeddings/bert-base-NER --warmup_epoch=0 --weight_decay=0.0 --epoch=20 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.9315955213435969, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 83032.81283378601ms, 22.508826101729316ms on average accuracy: 98.59%; precision: 92.93%; recall: 93.39%; FB1: 93.16 * --bert_model_name_or_path=./embedings/bert-base-cased --batch_size=32 --epoch=10 --use_crf INFO:__main__:[F1] : 0.8986090455778146, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 160266.587972641ms, 43.480443009338465ms on average accuracy: 97.95%; precision: 89.42%; recall: 90.93%; FB1: 90.17 * --bert_model_name_or_path=./embedings/bert-base-cased --batch_size=32 --epoch=10 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf INFO:__main__:[F1] : 0.9135758963967932, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 155832.32188224792ms, 42.27091953413272ms on average accuracy: 98.21%; precision: 91.30%; recall: 91.80%; FB1: 91.55 * --bert_model_name_or_path=./embedings/bert-base-cased --bert_disable_lstm --epoch=30 INFO:__main__:[F1] : 0.914522382798731, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 167704.35976982117ms, 45.50245252925677ms on average accuracy: 98.34%; precision: 91.04%; recall: 91.87%; FB1: 91.45 * using sub token label, --bert_use_sub_label --use_crf # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-base-cased --bert_use_sub_label # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=32 --lr=1e-5 --epoch=10 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf # evaluate $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint --use_crf $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9128322882628279, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 146032.53650665283ms, 39.613566708053355ms on average accuracy: 98.26%; precision: 91.59%; recall: 91.61%; FB1: 91.60 * using sub token label, --bert_use_sub_label + --bert_use_pos --use_char_cnn --use_crf # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-base-cased --bert_use_sub_label # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=32 --lr=1e-5 --epoch=10 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_pos --use_char_cnn # evaluate $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint --use_crf --bert_use_pos --use_char_cnn $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9113209212035649, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 148480.65185546875ms, 40.282283112823464ms on average accuracy: 98.27%; precision: 91.23%; recall: 91.52%; FB1: 91.37 ** using sub token label, --bert_use_sub_label + --bert_use_pos --use_char_cnn --epoch=30 --use_crf INFO:__main__:[F1] : 0.9142604856512141, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 146215.95406532288ms, 39.665775003881194ms on average accuracy: 98.32%; precision: 91.57%; recall: 91.75%; FB1: 91.66 * slicing logits to remain first token's of word's before applying crf, --use_crf --bert_use_crf_slice # https://github.com/dsindex/ntagger/releases/tag/v1.0 # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-base-cased # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=32 --lr=1e-5 --epoch=10 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_crf_slice # evaluate $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint --use_crf --bert_use_crf_slice $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.913277459197177, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 151587.14032173157ms, 41.12043155459907ms on average accuracy: 98.26%; precision: 91.01%; recall: 91.64%; FB1: 91.33 * subword pooling, --use_crf # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-base-cased --bert_use_subword_pooling # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=32 --lr=1e-5 --epoch=10 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_subword_pooling # evaluate $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint --use_crf --bert_use_subword_pooling INFO:__main__:[F1] : 0.9132734003172924, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 148011.45482063293ms, 40.14931384098779ms on average accuracy: 98.26%; precision: 90.93%; recall: 91.73%; FB1: 91.33 * subword pooling, word embedding, --use_crf # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-base-cased --bert_use_subword_pooling --bert_use_word_embedding # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=32 --lr=1e-5 --epoch=10 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_subword_pooling --bert_use_word_embedding # evaluate $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint --use_crf --bert_use_subword_pooling --bert_use_word_embedding INFO:__main__:[F1] : 0.9231446430143498, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 150338.47284317017ms, 40.779396660975905ms on average accuracy: 98.33%; precision: 91.81%; recall: 92.83%; FB1: 92.31 * subword pooling, word/pos/char embedding, --use_char_cnn --bert_use_pos --use_crf INFO:__main__:[F1] : 0.920799929521628, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 151703.0575275421ms, 41.146695921351785ms on average accuracy: 98.34%; precision: 91.64%; recall: 92.53%; FB1: 92.08 * subword pooling, --bert_disable_lstm, --batch_size=16 --epoch=30 INFO:__main__:[F1] : 0.9137430757056186, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 170345.35002708435ms, 46.214719198010975ms on average accuracy: 98.27%; precision: 90.76%; recall: 92.00%; FB1: 91.37 * subword pooling, word embedding, --bert_disable_lstm, --batch_size=16 --epoch=30 INFO:__main__:[F1] : 0.9120337790288531, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 122857.85698890686ms, 33.327639701733474ms on average accuracy: 98.28%; precision: 90.63%; recall: 91.78%; FB1: 91.20 * document context, --batch_size=16, n_ctx: 512, --epoch=30 --use_crf # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-base-cased --bert_use_doc_context # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=16 --lr=1e-5 --epoch=30 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_doc_context # evaluate $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint --use_crf --bert_use_doc_context $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9169225370646549, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 342276.90076828003ms, 92.88276947616009ms on average accuracy: 98.23%; precision: 91.56%; recall: 92.53%; FB1: 92.04 * document context, subword pooling, --batch_size=16, n_ctx: 512, --epoch=30 --use_crf $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=8 --lr=1e-5 --epoch=30 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_doc_context --bert_use_subword_pooling INFO:__main__:[F1] : 0.920799929521628, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 299821.12765312195ms, 81.3581446348051ms on average accuracy: 98.30%; precision: 91.64%; recall: 92.53%; FB1: 92.08 * document context, subword pooling, word embedding, --batch_size=8 --lr=1e-5, n_ctx: 512, --epoch=30 --use_crf $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=8 --lr=1e-5 --epoch=30 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_doc_context --bert_use_subword_pooling --bert_use_word_embedding INFO:__main__:[F1] : 0.9285462244177841, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 295369.29631233215ms, 80.1447755837939ms on average accuracy: 98.47%; precision: 92.53%; recall: 93.18%; FB1: 92.85 LOC: precision: 93.84%; recall: 94.06%; FB1: 93.95 1672 MISC: precision: 80.17%; recall: 82.34%; FB1: 81.24 721 ORG: precision: 91.42%; recall: 92.41%; FB1: 91.92 1679 PER: precision: 97.83%; recall: 97.77%; FB1: 97.80 1616 * document context, --bert_disable_lstm, --batch_size=16 --lr=2e-5, n_ctx: 512, --epoch=30 $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=16 --lr=2e-5 --epoch=30 --bert_use_doc_context --bert_disable_lstm INFO:__main__:[F1] : 0.9176470588235294, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 170834.2945575714ms, 46.35193099996301ms on average accuracy: 98.27%; precision: 91.01%; recall: 92.53%; FB1: 91.76 ** --lr=1e-5 INFO:__main__:[F1] : 0.9184963990865976, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 117725.76570510864ms, 31.933620964545387ms on average accuracy: 98.35%; precision: 91.13%; recall: 92.58%; FB1: 91.85 ** --bert_doc_context_option=2 # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-base-cased --bert_use_doc_context --bert_doc_context_option=2 INFO:__main__:[F1] : 0.9234554881799807, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 120362.42485046387ms, 32.65242268461071ms on average accuracy: 98.40%; precision: 91.68%; recall: 93.02%; FB1: 92.35 LOC: precision: 92.84%; recall: 94.00%; FB1: 93.42 1689 MISC: precision: 80.44%; recall: 83.19%; FB1: 81.79 726 ORG: precision: 89.92%; recall: 92.35%; FB1: 91.12 1706 PER: precision: 97.39%; recall: 96.97%; FB1: 97.18 1610 ** --bert_doc_context_option=2, try again! INFO:__main__:[F1] : 0.9232660228270413, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 101272.75466918945ms, 27.464767011196564ms on average accuracy: 98.44%; precision: 91.57%; recall: 93.09%; FB1: 92.33 * document context, subword pooling, --bert_disable_lstm, --batch_size=16 --lr=2e-5, n_ctx: 512, --epoch=30 $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=16 --lr=2e-5 --epoch=30 --bert_use_doc_context --bert_use_subword_pooling --bert_disable_lstm INFO:__main__:[F1] : 0.9223207232511191, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 172187.91961669922ms, 46.71668637358815ms on average accuracy: 98.41%; precision: 91.45%; recall: 93.02%; FB1: 92.23 ** --bert_doc_context_option=2 # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bert-base-cased --bert_use_doc_context --bert_use_subword_pooling --bert_doc_context_option=2 INFO:__main__:[F1] : 0.920534458509142, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 134746.63543701172ms, 36.54770698360927ms on average accuracy: 98.35%; precision: 91.41%; recall: 92.71%; FB1: 92.05 * document context, subword pooling, word embedding, --bert_disable_lstm, --batch_size=16 --lr=1e-5, n_ctx: 512, --epoch=30 $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=16 --lr=2e-5 --epoch=30 --bert_use_doc_context --bert_use_subword_pooling --bert_use_word_embedding --bert_disable_lstm INFO:__main__:[F1] : 0.9215065117916227, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 171518.04447174072ms, 46.53535477013941ms on average accuracy: 98.44%; precision: 91.60%; recall: 92.71%; FB1: 92.15 ** --bert_doc_context_option=2 --lr=1e-5 INFO:__main__:[F1] : 0.9202453987730062, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 173408.38193893433ms, 47.04242836286476ms on average accuracy: 98.36%; precision: 91.11%; recall: 92.95%; FB1: 92.02 * --bert_model_name_or_path=./embedings/bert-base-cased --batch_size=32 --epoch=10 INFO:__main__:[F1] : 0.9020433219328247, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 79621.61326408386ms, 21.58440276070066ms on average accuracy: 98.00%; precision: 89.37%; recall: 91.06%; FB1: 90.20 * --bert_model_name_or_path=./embedings/bert-base-cased --batch_size=32 --epoch=10 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.9098765432098764, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 80155.51614761353ms, 21.732826838828675ms on average accuracy: 98.24%; precision: 90.64%; recall: 91.34%; FB1: 90.99 * --bert_model_name_or_path=./embedings/bert-base-cased --batch_size=32 --epoch=10 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_mha INFO:__main__:[F1] : 0.9095074455899198, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 81080.08480072021ms, 21.98459193850871ms on average accuracy: 98.24%; precision: 90.53%; recall: 91.38%; FB1: 90.95 * --bert_model_name_or_path=./embedings/pytorch.uncased_L-8_H-512_A-8 INFO:__main__:[F1] : 0.8829307140960977, 3684 INFO:__main__:[Elapsed Time] : 99730ms, 27.048601683410265ms on average accuracy: 97.61%; precision: 88.25%; recall: 88.33%; FB1: 88.29 * --bert_model_name_or_path=./embedings/pytorch.uncased_L-4_H-512_A-8 INFO:__main__:[F1] : 0.8634692805881324, 3684 INFO:__main__:[Elapsed Time] : 83385ms, 22.60874287265816ms on average accuracy: 97.23%; precision: 85.38%; recall: 87.34%; FB1: 86.35 * --bert_model_name_or_path=./embedings/pytorch.uncased_L-4_H-256_A-4 INFO:__main__:[F1] : 0.8155101324677603, 3684 INFO:__main__:[Elapsed Time] : 79173ms, 21.463209340211783ms on average accuracy: 96.32%; precision: 80.82%; recall: 82.29%; FB1: 81.55 * --bert_model_name_or_path=./embedings/pytorch.uncased_L-2_H-128_A-2 INFO:__main__:[F1] : 0.6965218958370878, 3684 INFO:__main__:[Elapsed Time] : 74261ms, 20.137659516698342ms on average accuracy: 94.12%; precision: 70.92%; recall: 68.43%; FB1: 69.65 * --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/distilbert-base-uncased INFO:__main__:[F1] : 0.894963522897073, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 49652ms, 13.456421395601412ms on average accuracy: 97.80%; precision: 88.86%; recall: 90.14%; FB1: 89.50 * --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/distilbert-base-multilingual-cased INFO:__main__:[F1] : 0.9021052631578947, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 49763.64517211914ms, 13.491093125189254ms on average accuracy: 98.02%; precision: 89.39%; recall: 91.04%; FB1: 90.21 * --config=configs/config-bert.json --bert_model_name_or_path=./embeddings/MiniLM-L12-H384-uncased INFO:__main__:[F1] : 0.900193627882415, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 67914.6056175232ms, 18.416685349844798ms on average accuracy: 97.92%; precision: 89.50%; recall: 90.55%; FB1: 90.02 * --config=configs/config-bert.json --bert_model_name_or_path=./embeddings/MiniLM-L12-H384-uncased --warmup_epoch=0 --weight_decay=0.0 --epoch=30 INFO:__main__:[F1] : 0.9055271057743245, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 65600.49557685852ms, 17.789026203927158ms on average accuracy: 98.05%; precision: 90.31%; recall: 90.79%; FB1: 90.55 * --bert_model_name_or_path=./embeddings/bert-large-cased-whole-word-masking-finetuned-squad INFO:__main__:[F1] : 0.9130013221683562, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 130683ms, 35.44013032853652ms on average accuracy: 98.29%; precision: 90.91%; recall: 91.70%; FB1: 91.30 * --bert_model_name_or_path=./embeddings/bert-large-cased-whole-word-masking-finetuned-squad --warmup_epoch=0 --weight_decay=0.0 --epoch=30 INFO:__main__:[F1] : 0.9175393822054034, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 131497ms, 35.661960358403476ms on average accuracy: 98.33%; precision: 91.22%; recall: 92.30%; FB1: 91.75 * --data_dir=data/conll2003_truecase --bert_model_name_or_path=./embeddings/bert-large-cased-whole-word-masking-finetuned-squad --warmup_epoch=0 --weight_decay=0.0 --epoch=30 INFO:__main__:[F1] : 0.9217421785995968, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 121090.73448181152ms, 32.83806317143229ms on average accuracy: 98.32%; precision: 91.25%; recall: 93.11%; FB1: 92.17 * --bert_model_name_or_path=./embeddings/bert-large-cased-finetuned-conll03-english --warmup_epoch=0 --weight_decay=0.0 --epoch=30 INFO:__main__:[F1] : 0.9187279151943464, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 118809.88335609436ms, 32.22117830546304ms on average accuracy: 98.44%; precision: 91.68%; recall: 92.07%; FB1: 91.87 * --data_dir=data/conll2003_truecase --bert_model_name_or_path=./embeddings/bert-large-cased-finetuned-conll03-english --warmup_epoch=0 --weight_decay=0.0 --epoch=30 INFO:__main__:[F1] : 0.9239971850809289, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 114439.17107582092ms, 31.03302174406857ms on average accuracy: 98.44%; precision: 91.82%; recall: 92.99%; FB1: 92.40 * --bert_model_name_or_path=./embeddings/bert-large-cased-finetuned-conll03-english --warmup_epoch=0 --weight_decay=0.0 --epoch=30 --bert_disable_lstm INFO:__main__:[F1] : 0.9163143058491896, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 109043.2801246643ms, 29.56439605265443ms on average accuracy: 98.35%; precision: 91.18%; recall: 92.09%; FB1: 91.63 * --data_dir=data/conll2003_truecase --bert_model_name_or_path=./embeddings/bert-large-cased-finetuned-conll03-english --warmup_epoch=0 --weight_decay=0.0 --epoch=30 --bert_disable_lstm INFO:__main__:[F1] : 0.9224137931034484, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 110067.8985118866ms, 29.847684645827556ms on average accuracy: 98.41%; precision: 91.66%; recall: 92.83%; FB1: 92.24 * --data_dir=data/conll++ --bert_model_name_or_path=./embeddings/bert-large-cased-finetuned-conll03-english --warmup_epoch=0 --weight_decay=0.0 --epoch=30 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.926168632590651, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 132705.81126213074ms, 35.98520220410329ms on average accuracy: 98.52%; precision: 92.29%; recall: 92.95%; FB1: 92.62 * --data_dir=data/conll++_truecase --bert_model_name_or_path=./embeddings/bert-large-cased-finetuned-conll03-english --warmup_epoch=0 --weight_decay=0.0 --epoch=30 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.9336480461578809, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 132453.9577960968ms, 35.92103879930143ms on average accuracy: 98.63%; precision: 93.08%; recall: 93.65%; FB1: 93.36 * for using SpanBERT embedding, just replace pretrained BERT model to SpanBERT. * --bert_model_name_or_path=./embeddings/spanbert_hf_base INFO:__main__:[F1] : 0.9046450482033305, 3684 INFO:__main__:[Elapsed Time] : 110977ms, 30.09910399131143ms on average accuracy: 98.02%; precision: 89.57%; recall: 91.38%; FB1: 90.46 * --bert_model_name_or_path=./embeddings/spanbert_hf_large INFO:__main__:[F1] : 0.9139340659340659, 3684 INFO:__main__:[Elapsed Time] : 157069ms, 42.59598153679066ms on average accuracy: 98.23%; precision: 90.76%; recall: 92.03%; FB1: 91.39 * --data_dir=data/conll2003_truecase --bert_model_name_or_path=./embeddings/spanbert_hf_large --warmup_epoch=0 --weight_decay=0.0 --epoch=30 INFO:__main__:[F1] : 0.9201266713581985, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 116231.20188713074ms, 31.509835921387747ms on average accuracy: 98.35%; precision: 91.43%; recall: 92.60%; FB1: 92.01 * --bert_use_feature_based --epoch=64 --lr=1e-4 , modify model.py to use initial embedding INFO:__main__:[F1] : 0.8610818405338954, 3684 INFO:__main__:[Elapsed Time] : 181867ms, 49.310344827586206ms on average accuracy: 97.43%; precision: 85.42%; recall: 86.81%; FB1: 86.11 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use initial embedding INFO:__main__:[F1] : 0.8575426322693486, 3684 INFO:__main__:[Elapsed Time] : 232758ms, 63.13765951669834ms on average accuracy: 97.06%; precision: 86.06%; recall: 86.81%; FB1: 86.43 * --bert_use_feature_based --epoch=64 --lr=3e-4 , modify model.py to use initial+first+last embedding INFO:__main__:[F1] : 0.8971681415929202, 3684 INFO:__main__:[Elapsed Time] : 176851ms, 47.970404561498775ms on average accuracy: 98.00%; precision: 89.69%; recall: 89.75%; FB1: 89.72 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use initial+first+last embedding INFO:__main__:[F1] : 0.8946078431372549, 3684 INFO:__main__:[Elapsed Time] : 247716ms, 67.20418137387999ms on average accuracy: 97.81%; precision: 89.46%; recall: 90.47%; FB1: 89.96 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use last embedding INFO:__main__:[F1] : 0.8912471496228732, 3684 INFO:__main__:[Elapsed Time] : 253449ms, 68.75481944067336ms on average accuracy: 97.75%; precision: 89.38%; recall: 89.96%; FB1: 89.67 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use [-4:] embedding INFO:__main__:[F1] : 0.90263319044703, 3684 INFO:__main__:[Elapsed Time] : 235691ms, 63.9397230518599ms on average accuracy: 97.99%; precision: 89.94%; recall: 91.34%; FB1: 90.64 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use mean([0:3] + [-4:]) embedding INFO:__main__:[F1] : 0.9026813186813187, 3684 INFO:__main__:[Elapsed Time] : 261161ms, 70.83220200923161ms on average accuracy: 98.02%; precision: 90.15%; recall: 90.90%; FB1: 90.52 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use mean([0:17]) embedding INFO:__main__:[F1] : 0.9050141242937852, 3684 INFO:__main__:[Elapsed Time] : 253013ms, 68.61390171056205ms on average accuracy: 98.06%; precision: 90.87%; recall: 90.76%; FB1: 90.81 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use max([0:17]) embedding INFO:__main__:[F1] : 0.9054972205064854, 3684 INFO:__main__:[Elapsed Time] : 224124ms, 60.80396415965246ms on average accuracy: 98.09%; precision: 90.67%; recall: 90.85%; FB1: 90.76 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use mean([0:]) embedding INFO:__main__:[F1] : 0.9071163694155041, 3684 INFO:__main__:[Elapsed Time] : 217202ms, 58.91121368449633ms on average accuracy: 98.13%; precision: 91.00%; recall: 90.95%; FB1: 90.98 * --bert_use_feature_based --use_crf --epoch=64 --lr=3e-4 , modify model.py to use DSA(4, 300) INFO:__main__:[F1] : 0.9036879808967896, 3684 INFO:__main__:[Elapsed Time] : 245729ms, 66.65761607385284ms on average accuracy: 98.08%; precision: 90.78%; recall: 90.46%; FB1: 90.62 * --bert_use_feature_based --use_crf --epoch=64 --lr=1e-3 , modify model.py to use DSA(2, 1024) INFO:__main__:[F1] : 0.8953335090957026, 3684 INFO:__main__:[Elapsed Time] : 219016ms, 59.40619060548466ms on average accuracy: 97.94%; precision: 89.37%; recall: 90.19%; FB1: 89.78 * --bert_model_name_or_path=./embedings/bert-base-uncased --bert_remove_layers=8,9,10,11 INFO:__main__:[F1] : 0.8902760682257781, 3684 INFO:__main__:[Elapsed Time] : 91854ms, 24.907683953298942ms on average accuracy: 97.75%; precision: 88.42%; recall: 89.64%; FB1: 89.03 * --bert_remove_layers=12,13,14,15,16,17,18,19,20,21,22,23 INFO:__main__:[F1] : 0.8910284463894966, 3684 INFO:__main__:[Elapsed Time] : 122182ms, 33.13765951669834ms on average accuracy: 97.80%; precision: 88.11%; recall: 90.12%; FB1: 89.10 ```

emb_class=albert, enc_class=bilstm

emb_class=roberta, enc_class=bilstm

- train ``` * n_ctx size should be less than 512 $ python preprocess.py --config=configs/config-roberta.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/roberta-large $ python train.py --config=configs/config-roberta.json --data_dir=data/conll2003 --save_path=pytorch-model-roberta.pt --bert_model_name_or_path=./embeddings/roberta-large --bert_output_dir=bert-checkpoint-roberta --batch_size=16 --lr=1e-5 --epoch=10 ``` - evaluation ``` $ python evaluate.py --config=configs/config-roberta.json --data_dir=data/conll2003 --model_path=pytorch-model-roberta.pt --bert_output_dir=bert-checkpoint-roberta $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9119915848527349, 3684 INFO:__main__:[Elapsed Time] : 145234ms, 39.37578061363019ms on average accuracy: 98.27%; precision: 90.31%; recall: 92.10%; FB1: 91.20 * --batch_size=32 INFO:__main__:[F1] : 0.9182367890631846, 3684 INFO:__main__:[Elapsed Time] : 147221ms, 39.92967689383654ms on average * --bert_disable_lstm INFO:__main__:[F1] : 0.9183817062445031, 3684 F INFO:__main__:[Elapsed Time] : 105306ms, 28.55253869128428ms on average * --use_crf INFO:__main__:[F1] : 0.9013054830287206, 3684 INFO:__main__:[Elapsed Time] : 221208ms, 60.01221830029867ms on average * --bert_use_pos INFO:__main__:[F1] : 0.914000175330937, 3684 INFO:__main__:[Elapsed Time] : 153930ms, 41.748574531631824ms on average * --data_dir=data/conll2003_truecase --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --epoch=30 INFO:__main__:[F1] : 0.9190283400809718, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 112529.90531921387ms, 30.516853094295158ms on average accuracy: 98.30%; precision: 91.37%; recall: 92.44%; FB1: 91.90 * --bert_model_name_or_path=./embeddings/roberta-base --bert_disable_lstm INFO:__main__:[F1] : 0.9002973587545915, 3684 INFO:__main__:[Elapsed Time] : 71015ms, 19.25033939723052ms on average * --bert_model_name_or_path=./embeddings/xlm-roberta-base --bert_disable_lstm --batch_size=32 --epoch=30 --patience=4 INFO:__main__:[F1] : 0.912046393111326, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 69937.09135055542ms, 18.960459090560924ms on average accuracy: 98.21%; precision: 90.53%; recall: 91.89%; FB1: 91.20 * --bert_model_name_or_path=./embeddings/xlm-roberta-base --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --batch_size=32 --epoch=30 --patience=4 INFO:__main__:[F1] : 0.9081166549543219, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 79163.73658180237ms, 21.466715388281205ms on average accuracy: 98.15%; precision: 90.12%; recall: 91.52%; FB1: 90.81 * --bert_model_name_or_path=./embeddings/xlm-roberta-base --use_crf --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --batch_size=32 --epoch=30 --patience=4 INFO:__main__:[F1] : 0.9111501316944689, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 145396.85487747192ms, 39.44182609560177ms on average accuracy: 98.17%; precision: 90.37%; recall: 91.87%; FB1: 91.12 * using sub token label, --bert_use_sub_label # preprocessing $ python preprocess.py --config=configs/config-roberta.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/xlm-roberta-base --bert_use_sub_label # --bert_model_name_or_path=./embeddings/xlm-roberta-base --use_crf --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --batch_size=32 --epoch=30 --patience=4 INFO:__main__:[F1] : 0.9172601776136463, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 158769.8678970337ms, 43.06626947813632ms on average accuracy: 98.28%; precision: 91.24%; recall: 92.35%; FB1: 91.79 * slicing logits, --bert_use_crf_slice # https://github.com/dsindex/ntagger/releases/tag/v1.0 # train $ python train.py --config=configs/config-roberta.json --data_dir=data/conll2003 --save_path=pytorch-model-roberta.pt --bert_model_name_or_path=./embeddings/xlm-roberta-base --bert_output_dir=bert-checkpoint-roberta --batch_size=32 --lr=1e-5 --epoch=30 --patience=4 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_crf_slice INFO:__main__:[F1] : 0.911604155661208, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 145113.0769252777ms, 39.364287298463005ms on average accuracy: 98.20%; precision: 90.67%; recall: 91.66%; FB1: 91.16 * --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_disable_lstm INFO:__main__:[F1] : 0.927465220054248, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 111304.85033988953ms, 30.18471685208091ms on average INFO:__main__:[Elapsed Time] : 100 examples, 2893.0163383483887ms, 27.91449758741591ms on average accuracy: 98.47%; precision: 91.68%; recall: 93.84%; FB1: 92.75 * --bert_model_name_or_path=./embeddings/xlm-roberta-large --data_dir=data/conll2003_truecase --bert_disable_lstm INFO:__main__:[F1] : 0.928909952606635, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 112505.32031059265ms, 30.506469942518073ms on average INFO:__main__:[Elapsed Time] : 100 examples, 2932.1699142456055ms, 28.066835018119427ms on average accuracy: 98.48%; precision: 92.10%; recall: 93.70%; FB1: 92.89 * subword pooling, word embedding # preprocessing $ python preprocess.py --config=configs/config-roberta.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_use_subword_pooling --bert_use_word_embedding # train $ python train.py --config=configs/config-roberta.json --data_dir=data/conll2003 --save_path=pytorch-model-roberta.pt --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_output_dir=bert-checkpoint-roberta --batch_size=32 --lr=1e-5 --epoch=30 --patience=4 --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 --use_crf --bert_use_subword_pooling --bert_use_word_embedding # evaluate $ python evaluate.py --config=configs/config-roberta.json --data_dir=data/conll2003 --model_path=pytorch-model-roberta.pt --bert_output_dir=bert-checkpoint-roberta --use_crf --bert_use_subword_pooling --bert_use_word_embedding INFO:__main__:[F1] : 0.9296765119549929, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 194712.32175827026ms, 52.813371497264804ms on average accuracy: 98.44%; precision: 92.32%; recall: 93.63%; FB1: 92.97 * --data_dir=data/conll++ --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_disable_lstm INFO:__main__:[F1] : 0.9394968224949942, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 123247.82752990723ms, 33.42233563841862ms on average accuracy: 98.69%; precision: 93.28%; recall: 94.63%; FB1: 93.95 * --data_dir=data/conll++_truecase --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_disable_lstm INFO:__main__:[F1] : 0.941063308849248, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 121127.03013420105ms, 32.838388593695974ms on average accuracy: 98.75%; precision: 93.84%; recall: 94.37%; FB1: 94.11 * --data_dir=data/conll++ --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.9374890867819102, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 127180.99355697632ms, 34.48944825890464ms on average accuracy: 98.69%; precision: 93.34%; recall: 94.16%; FB1: 93.75 * --data_dir=data/conll++_truecase --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_freezing_epoch=3 --bert_lr_during_freezing=1e-3 INFO:__main__:[F1] : 0.9381488266596877, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 127575.44207572937ms, 34.58970893571922ms on average accuracy: 98.65%; precision: 93.33%; recall: 94.30%; FB1: 93.81 * document context, --bert_disable_lstm --batch_size=8, n_ctx: 512, --epoch=30 # preprocessing $ python preprocess.py --config=configs/config-roberta.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_use_doc_context --bert_doc_context_option=2 # train $ python train.py --config=configs/config-roberta.json --data_dir=data/conll2003 --save_path=pytorch-model-roberta.pt --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_output_dir=bert-checkpoint-roberta --batch_size=8 --lr=1e-5 --epoch=30 --bert_use_doc_context --bert_disable_lstm --eval_batch_size=32 # evaluate $ python evaluate.py --config=configs/config-roberta.json --data_dir=data/conll2003 --model_path=pytorch-model-roberta.pt --bert_output_dir=bert-checkpoint-roberta --bert_use_doc_context --bert_disable_lstm $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.938635161347205, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 185295.64094543457ms, 50.268435588259756ms on average accuracy: 98.73%; precision: 93.48%; recall: 94.25%; FB1: 93.86 LOC: precision: 95.44%; recall: 94.12%; FB1: 94.78 1645 MISC: precision: 82.48%; recall: 87.18%; FB1: 84.76 742 ORG: precision: 92.06%; recall: 93.56%; FB1: 92.80 1688 PER: precision: 98.02%; recall: 98.14%; FB1: 98.08 1619 * document context, subword pooling, word embedding, --batch_size=8, n_ctx: 512, --epoch=30 --use_crf # preprocessing $ python preprocess.py --config=configs/config-roberta.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_use_doc_context --bert_doc_context_option=2 --bert_use_subword_pooling --bert_use_word_embedding # train $ python train.py --config=configs/config-roberta.json --data_dir=data/conll2003 --save_path=pytorch-model-roberta.pt --bert_model_name_or_path=./embeddings/xlm-roberta-large --bert_output_dir=bert-checkpoint-roberta --batch_size=8 --lr=1e-5 --epoch=30 --bert_use_doc_context --bert_use_subword_pooling --bert_use_word_embedding --use_crf --eval_batch_size=32 --patience=4 --bert_freezing_epoch=1 --bert_lr_during_freezing=1e-3 --eval_steps=1000 # evaluate $ python evaluate.py --config=configs/config-roberta.json --data_dir=data/conll2003 --model_path=pytorch-model-roberta.pt --bert_output_dir=bert-checkpoint-roberta --bert_use_doc_context --bert_use_subword_pooling --bert_use_word_embedding --use_crf $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9358929200422684, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 387618.5405254364ms, 105.18818713386779ms on average accuracy: 98.61%; precision: 93.10%; recall: 94.09%; FB1: 93.59 ```

emb_class=bart, enc_class=bilstm

- train ``` * n_ctx size should be less than 512 $ python preprocess.py --config=configs/config-bart.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/bart-large $ python train.py --config=configs/config-bart.json --data_dir=data/conll2003 --save_path=pytorch-model-bart.pt --bert_model_name_or_path=./embeddings/bart-large --bert_output_dir=bert-checkpoint-bart --batch_size=32 --lr=1e-5 --epoch=10 ``` - evaluation ``` $ python evaluate.py --config=configs/config-bart.json --data_dir=data/conll2003 --model_path=pytorch-model-bart.pt --bert_output_dir=bert-checkpoint-bart $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9042961928047503, 3684 INFO:__main__:[Elapsed Time] : 196742ms, 53.36573445560684ms on average accuracy: 98.04%; precision: 89.21%; recall: 91.68%; FB1: 90.43 ```

emb_class=electra, enc_class=bilstm

emb_class=deberta, enc_class=bilstm

- train ``` * share config-bert.json * n_ctx size should be less than 512 $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/deberta-base $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-deberta.pt --bert_model_name_or_path=./embeddings/deberta-base --bert_output_dir=bert-checkpoint-deberta --batch_size=32 --lr=1e-5 --epoch=20 --bert_disable_lstm ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-deberta.pt --bert_output_dir=bert-checkpoint-deberta --bert_disable_lstm $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9040613161835961, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 114359.83681678772ms, 31.012554386210383ms on average accuracy: 98.10%; precision: 89.97%; recall: 90.85%; FB1: 90.41 INFO:__main__:[Elapsed Time] : 100 examples, 2987.0100021362305ms, 28.687477111816406ms on average * --bert_model_name_or_path=./embeddings/deberta-large --batch_size=16 --gradient_accumulation_steps=2 INFO:__main__:[F1] : 0.9144771645212484, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 206926.0606765747ms, 56.128300501864594ms on average INFO:__main__:[Elapsed Time] : 100 examples, 5544.259309768677ms, 53.92497717732131ms on average accuracy: 98.32%; precision: 91.08%; recall: 91.82%; FB1: 91.45 * --use_sharded_ddp # https://github.com/dsindex/ntagger/releases/tag/v1.1 $ export NCCL_DEBUG=INFO $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-deberta.pt --bert_model_name_or_path=./embeddings/deberta-v2-xlarge/ --bert_output_dir=bert-checkpoint-deberta --batch_size=16 --lr=1e-5 --epoch=20 --bert_disable_lstm --gradient_accumulation_steps=2 --use_sharded_ddp --world_size=2 --master_port=8666 INFO:__main__:[F1] : 0.931234611326064, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 232110.55159568787ms, 62.972230753301766ms on average accuracy: 98.57%; precision: 92.51%; recall: 93.75%; FB1: 93.12 * --use_sharded_ddp --use_fsdp # https://github.com/dsindex/ntagger/releases/tag/v1.1 $ export NCCL_DEBUG=INFO $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-deberta.pt --bert_model_name_or_path=./embeddings/deberta-v2-xlarge/ --bert_output_dir=bert-checkpoint-deberta --batch_size=16 --lr=1e-5 --epoch=20 --bert_disable_lstm --gradient_accumulation_steps=2 --use_sharded_ddp --use_fsdp --world_size=2 --master_port=5176 # 0 accuracy! how to load the saved model from FSDP.state_dict()? * --use_sharded_ddp, document context, --bert_disable_lstm --batch_size=8, n_ctx: 512, --epoch=20 # https://github.com/dsindex/ntagger/releases/tag/v1.1 # preprocessing $ python preprocess.py --config=configs/config-bert.json --data_dir=data/conll2003 --bert_model_name_or_path=./embeddings/deberta-v2-xlarge --bert_use_doc_context --bert_doc_context_option=2 # train $ python train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-deberta.pt --bert_model_name_or_path=./embeddings/deberta-v2-xlarge --bert_output_dir=bert-checkpoint-deberta --batch_size=4 --gradient_accumulation_steps=4 --lr=1e-5 --epoch=20 --bert_use_doc_context --bert_disable_lstm --eval_batch_size=16 --use_sharded_ddp --world_size=2 --master_port=20630 # evaluate $ python evaluate.py --config=configs/config-bert.json --data_dir=data/conll2003 --model_path=pytorch-model-deberta.pt --bert_output_dir=bert-checkpoint-deberta --bert_use_doc_context --bert_disable_lstm $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9400386847195358, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 578826.6246318817ms, 157.07525959682957ms on average accuracy: 98.73%; precision: 93.36%; recall: 94.65%; FB1: 94.00 LOC: precision: 94.47%; recall: 95.26%; FB1: 94.87 1682 MISC: precision: 80.48%; recall: 86.32%; FB1: 83.30 753 ORG: precision: 92.81%; recall: 94.10%; FB1: 93.45 1684 PER: precision: 98.82%; recall: 98.21%; FB1: 98.51 1607 * accelerate & torch.distributed.launch # 1 node, 2 gpu $ export NCCL_DEBUG=INFO $ python -m torch.distributed.launch --nnodes 1 --nproc_per_node 2 --use_env --node_rank 0 --master_addr 10.55.14.139 --master_port 3667 train.py --config=configs/config-bert.json --data_dir=data/conll2003 --save_path=pytorch-model-deberta.pt --bert_model_name_or_path=./embeddings/deberta-v2-xlarge/ --bert_output_dir=bert-checkpoint-deberta --batch_size=8 --lr=1e-5 --epoch=20 --bert_disable_lstm --gradient_accumulation_steps=4 --eval_batch_size=4 INFO:__main__:[F1] : 0.9230634266163698, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 233286.6771221161ms, 63.29245888406804ms on average ```

emb_class=elmo, enc_class=bilstm

- train ``` * token_emb_dim in configs/config-elmo.json == 300 (ex, glove.6B.300d.txt ) * elmo_emb_dim in configs/config-elmo.json == 1024 (ex, elmo_2x4096_512_2048cnn_2xhighway_5.5B_* ) $ python preprocess.py --config=configs/config-elmo.json --data_dir=data/conll2003 --embedding_path=embeddings/glove.6B.300d.txt * --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding $ python train.py --config=configs/config-elmo.json --data_dir=data/conll2003 --save_path=pytorch-model-elmo.pt --use_crf ``` - evaluation ``` $ python evaluate.py --config=configs/config-elmo.json --data_dir=data/conll2003 --model_path=pytorch-model-elmo.pt --use_crf $ cd data/conll2003; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.9219494967331803, 3684 INFO:__main__:[Elapsed Time] : 239919ms, 65.12459283387622ms on average accuracy: 98.29%; precision: 91.95%; recall: 92.44%; FB1: 92.19 * --batch_size=64 INFO:__main__:[F1] : 0.926332565964229, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 275097ms, 74.65218571816455ms on average accuracy: 98.45%; precision: 92.65%; recall: 92.62%; FB1: 92.63 * --data_dir=data/conll2003_truecase --batch_size=64 --warmup_epoch=0 --weight_decay=0.0 INFO:__main__:[F1] : 0.9251124636147129, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 236348.47974777222ms, 64.13688945796157ms on average accuracy: 98.40%; precision: 92.18%; recall: 92.85%; FB1: 92.51 * --use_char_cnn INFO:__main__:[F1] : 0.9137136782423814, 3684 INFO:__main__:[Elapsed Time] : 280253ms, 76.05566114580505ms on average accuracy: 98.15%; precision: 91.44%; recall: 91.31%; FB1: 91.37 * --use_char_cnn INFO:__main__:[F1] : 0.9196578181497487, 3684 INFO:__main__:[Elapsed Time] : 291372ms, 79.06489275047515ms on average accuracy: 98.26%; precision: 91.62%; recall: 92.32%; FB1: 91.97 * --use_char_cnn --batch_size=64 INFO:__main__:[F1] : 0.9202508169213106, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 222863ms, 60.46673907140918ms on average accuracy: 98.32%; precision: 91.81%; recall: 92.25%; FB1: 92.03 * modify model.py for disabling glove INFO:__main__:[F1] : 0.917751217352811, 3684 INFO:__main__:[Elapsed Time] : 273089ms, 74.10019006244909ms on average accuracy: 98.27%; precision: 91.78%; recall: 91.77%; FB1: 91.78 * --use_char_cnn , modify model.py for disabling glove INFO:__main__:[F1] : 0.9193262411347518, 3684 INFO:__main__:[Elapsed Time] : 249467ms, 67.69318490361118ms on average accuracy: 98.31%; precision: 92.06%; recall: 91.80%; FB1: 91.93 * --data_dir=data/conll++ --batch_size=64 INFO:__main__:[F1] : 0.9349278930706999, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 240186.9957447052ms, 65.17468371405538ms on average accuracy: 98.59%; precision: 93.76%; recall: 93.23%; FB1: 93.49 * --data_dir=data/conll++_truecase --batch_size=64 INFO:__main__:[F1] : 0.9368402533427164, 3684 INFO:__main__:[Elapsed Time] : 3684 examples, 245365.75412750244ms, 66.58202281892866ms on average accuracy: 98.64%; precision: 93.98%; recall: 93.39%; FB1: 93.68 ```

Kaggle NER (English)

experiments summary

ntagger, measured by conlleval.pl (micro F1)

	F1 (%)	Features	GPU / CPU	CONDA	ONNX	Dynamic
GloVe, BiLSTM-CRF	85.67	word, pos	23.7084 / -
GloVe, BiLSTM-CRF	85.78	word, character, pos	24.2101 / -
BERT-base(cased), BiLSTM	84.43	word	39.0914 / -	-	-	-
BERT-base(cased), BiLSTM-CRF	84.62	word, pos	37.7312 / -	-	-	-
BERT-large-squad, BiLSTM-CRF	84.78	word, pos	53.3669 / -
ELMo, GloVe, BiLSTM-CRF	85.48	word, pos	80.5333 / -	-

emb_class=glove, enc_class=bilstm

- train ``` * token_emb_dim in configs/config-glove.json == 300 (ex, glove.6B.300d.txt ) $ python preprocess.py --data_dir=data/kaggle * --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding $ python train.py --data_dir=data/kaggle --use_crf ``` - evaluation ``` $ python evaluate.py --data_dir=data/kaggle --use_crf $ cd data/kaggle; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8567375886524823, 4795 INFO:__main__:[Elapsed Time] : 4795 examples, 113756.6728591919ms, 23.708498821091442ms on average accuracy: 97.38%; precision: 85.79%; recall: 85.56%; FB1: 85.67 * --use_char_cnn INFO:__main__:[F1] : 0.8578012141622723, 4795 INFO:__main__:[Elapsed Time] : 4795 examples, 116158.74361991882ms, 24.210153393910534ms on average accuracy: 97.38%; precision: 85.87%; recall: 85.69%; FB1: 85.78 ```

emb_class=bert, enc_class=bilstm

- train ``` * n_ctx size should be less than 512 $ python preprocess.py --config=configs/config-bert.json --data_dir=data/kaggle --bert_model_name_or_path=./embeddings/bert-base-cased $ python train.py --config=configs/config-bert.json --data_dir=data/kaggle --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=32 --lr=5e-5 --epoch=20 --warmup_epoch=0 --weight_decay=0.0 --patience=4 ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --data_dir=data/kaggle --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint $ cd data/kaggle; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8443369254453487, 4795 INFO:__main__:[Elapsed Time] : 4795 examples, 187503.9415359497ms, 39.09142493803799ms on average * {'lr': 5.2929204434436896e-05, 'batch_size': 32, 'seed': 27, 'epochs': 15} by optuna INFO:__main__:[F1] : 0.840452206044077, 4795 INFO:__main__:[Elapsed Time] : 4795 examples, 185833.47272872925ms, 38.74066833858944ms on average accuracy: 97.17%; precision: 83.52%; recall: 84.58%; FB1: 84.05 * --bert_use_pos --use_crf INFO:__main__:[F1] : 0.8459812590735118, 4795 INFO:__main__:[Elapsed Time] : 4795 examples, 181019.96636390686ms, 37.73124167259703ms on average accuracy: 97.31%; precision: 84.11%; recall: 85.13%; FB1: 84.62 * --bert_model_name_or_path=./embeddings/bert-large-cased-whole-word-masking-finetuned-squad --bert_use_pos --use_crf INFO:__main__:[F1] : 0.8475636619966518, 4795 INFO:__main__:[Elapsed Time] : 4795 examples, 256007.39908218384ms, 53.3669921921152ms on average accuracy: 97.30%; precision: 84.40%; recall: 85.16%; FB1: 84.78 ```

emb_class=elmo, enc_class=bilstm

GUM (English)

experiments summary

ntagger, measured by conlleval.pl (micro F1)

	F1 (%)	Features	GPU / CPU	CONDA	ONNX	Dynamic
GloVe, BiLSTM-CRF	53.70	word, character	24.3435 / -
BERT-base(cased), BiLSTM-CRF	63.13	word	40.3382 / -	-	-	-
BERT-large-squad, BiLSTM-CRF	62.76	word	63.1012 / -
ELMo, GloVe, BiLSTM-CRF	61.46	word	79.0402 / -	-

emb_class=glove, enc_class=bilstm

- train ``` * token_emb_dim in configs/config-glove.json == 300 (ex, glove.6B.300d.txt ) $ python preprocess.py --data_dir=data/gum * --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding $ python train.py --data_dir=data/gum --use_crf --use_char_cnn ``` - evaluation ``` $ python evaluate.py --data_dir=data/gum --use_crf --use_char_cnn $ cd data/gum; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.5370499419279907, 1000 INFO:__main__:[Elapsed Time] : 1000 examples, 24429.93712425232ms, 24.343522103341133ms on average accuracy: 79.30%; precision: 55.47%; recall: 52.05%; FB1: 53.70 ```

emb_class=bert, enc_class=bilstm

- train ``` * n_ctx size should be less than 512 $ python preprocess.py --config=configs/config-bert.json --data_dir=data/gum --bert_model_name_or_path=./embeddings/bert-base-cased $ python train.py --config=configs/config-bert.json --data_dir=data/gum --save_path=pytorch-model-bert.pt --bert_model_name_or_path=./embeddings/bert-base-cased --bert_output_dir=bert-checkpoint --batch_size=32 --lr=5e-5 --epoch=20 --warmup_epoch=0 --weight_decay=0.0 --patience=4 --use_crf ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --data_dir=data/gum --model_path=pytorch-model-bert.pt --bert_output_dir=bert-checkpoint --use_crf $ cd data/gum; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. * {'lr': 5.8243506260824845e-05, 'batch_size': 32, 'seed': 42, 'epochs': 18} by optuna INFO:__main__:[F1] : 0.6304728546409808, 1000 INFO:__main__:[Elapsed Time] : 1000 examples, 40466.28999710083ms, 40.33822578949494ms on average accuracy: 83.86%; precision: 61.51%; recall: 64.84%; FB1: 63.13 * --bert_model_name_or_path=./embeddings/bert-large-cased-whole-word-masking-finetuned-squad INFO:__main__:[F1] : 0.6267972494269639, 1000 INFO:__main__:[Elapsed Time] : 1000 examples, 63216.368436813354ms, 63.10122626441139ms on average accuracy: 82.62%; precision: 58.48%; recall: 67.72%; FB1: 62.76 ```

emb_class=elmo, enc_class=bilstm

Naver NER 2019 (Korean)

experiments summary

clova2019(eoj-based)

ntagger, measured by conlleval.pl (micro F1)

	F1 (%)	Features	GPU / CPU	Etc
bpe DistilBERT(v1)	85.30	eoj	9.0702 / -
wp DistilBERT(v1)	84.45	eoj	8.9646 / -
mDistilBERT	83.89	eoj	9.2205 / -
bpe BERT(v1), BiLSTM-CRF	86.34	eoj	46.9389 / -
bpe BERT(v1), BiLSTM-CRF	87.17	eoj	39.1787 / -	freezing BERT during some epochs
bpe BERT(v1), BiLSTM-CRF	86.99	eoj	39.0575 / -	using sub token label, freezing BERT during some epochs
bpe BERT(v1), BiLSTM-CRF	87.46	eoj	39.6914 / -	slicing logits, freezing BERT during some epochs, https://github.com/dsindex/ntagger/releases/tag/v1.0
bpe BERT(v1), BiLSTM	86.37	eoj	21.3232 / -
bpe BERT(v1), CRF	86.42	eoj	35.2222 / -
bpe BERT(v1)	87.13	eoj	16.2121 / -
bpe BERT-large(v1)	85.99	eoj	30.7513 / -
bpe BERT-large(v1), BiLSTM	85.82	eoj	32.0083 / -	freezing BERT during some epochs
bpe BERT-large(v3)	85.89	eoj	27.4264 / -
MA-BERT-base	86.40	eoj	13.3906 / -
KcBERT-base, BiLSTM	84.76	eoj	15.0553 / -
KcBERT-base, CRF	83.32	eoj	31.8019 / -
KcBERT-base	84.72	eoj	13.3129 / -
KcBERT-large	86.34	eoj	26.9639 / -
KoELECTRA-Base-v1	86.64	eoj	15.1616 / -
KoELECTRA-Base-v3	87.31	eoj	14.8115 / -
KoELECTRA-Base-v3, BiLSTM-CRF	87.76	eoj	40.4698 / -	freezing BERT during some epochs
KoELECTRA-Base-v3, BiLSTM-CRF	87.32	eoj	39.8039 / -	using sub token label, freezing BERT during some epochs
KoELECTRA-Base-v3, BiLSTM-CRF	88.13	eoj	40.0855 / -	slicing logits, freezing BERT during some epochs, https://github.com/dsindex/ntagger/releases/tag/v1.0
KoELECTRA-Base-v3, BiLSTM-CRF	87.89	eoj	47.0637 / -	subword pooling
LM-KOR-ELECTRA	87.39	eoj	17.1545 / -
LM-KOR-ELECTRA, BiLSTM-CRF	87.49	eoj	39.7247 / -	slicing logits, freezing BERT during some epochs, https://github.com/dsindex/ntagger/releases/tag/v1.0
bpe ELECTRA-base(v1)	86.46	eoj	18.0449 / -
dhaToken1.large ELECTRA-base, BiLSTM-CRF	86.90	eoj	44.3714 / -	slicing logits, freezing BERT during some epochs, https://github.com/dsindex/ntagger/releases/tag/v1.0
dhaSyllable ELECTRA-base, BiLSTM-CRF	86.31	eoj	41.0562 / -	subword pooling
RoBERTa-base	85.45	eoj	15.6986 / -
MA-RoBERTa-base	87.80	eoj	15.7000 / -
XLM-RoBERTa-base	86.84	eoj	18.1326 / -
XLM-RoBERTa-large	87.01	eoj	35.9521 / -
KLUE-RoBERTa-base	87.74	eoj	17.3930 / -
KLUE-RoBERTa-base	87.83	eoj	22.3236 / -	subword pooling
KLUE-RoBERTa-large	88.75	eoj	30.9014 / -
KLUE-RoBERTa-large	88.74	eoj	33.8022 / -	subword pooling
Funnel-base	87.97	eoj	42.9287 / -
Funnel-base, BiLSTM-CRF	87.92	eoj	83.9707 / -	slicing logits, freezing BERT during some epochs, https://github.com/dsindex/ntagger/releases/tag/v1.0

HanBert-NER, KoELECTRA, LM-kor measured by seqeval (micro F1)

(update) max_seq_len=50	F1 (%)	Features
BiLSTM-CRF	74.57	eoj
Bert-multilingual	84.20	eoj
DistilKoBERT	84.13	eoj
KoBERT	87.92	eoj
HanBert	87.70	eoj
KoELECTRA-Base	87.18	eoj
KoELECTRA-Base-v2	87.16	eoj
KoELECTRA-Base-v3	88.11	eoj
electra-kor-base	87.14	eoj
funnel-kor-base	88.02	eoj

* note that F1 score from the 'seqeval' package for 'max_seq_len=50' might be similar with that for 'max_seq_len=180'. 
  however, the final evaluation using 'conlleval.pl' should be different.

  for example, with n_ctx=50. 
    the F1 score from 'seqeval' is 0.8602. 
    the F1 score from 'conlleval.pl is 0.8438.
  --------------------------------------------------------------------------
  '--bert_disable_lstm ,  n_ctx=50'
  INFO:__main__:[F1] : 0.8602524268436113, 9000
  INFO:__main__:[Elapsed Time] : 192653ms, 21.39648849872208ms on average
  accuracy:  93.22%; precision:  85.55%; recall:  83.25%; FB1:  84.38

  '--bert_disable_lstm , without --use_crf (bpe BERT), n_ctx=180'
  INFO:__main__:[F1] : 0.8677214324767633, 9000
  INFO:__main__:[Elapsed Time] : 868094ms, 96.45471719079897ms on average
  accuracy:  94.47%; precision:  87.02%; recall:  86.33%; FB1:  86.68
  --------------------------------------------------------------------------

  this is due to the test.txt data has longer sequences than the 'n_ctx' size.
  therefore, the evaluation results using 'seqeval' are not the final F1 score.
  we recommend to use 'conlleval.pl' script for NER results.

clova2019_morph(morph-based)

ntagger, measured by conlleval.pl (micro F1)

	m-by-m F1 (%)	e-by-e F1 (%)	Features	GPU / CPU	Etc
GloVe, BiLSTM-CRF	84.29	84.29	morph, pos	30.0968 / -
GloVe, BiLSTM-CRF	85.82	85.82	morph, character, pos	25.9623 / -
GloVe, DenseNet-CRF	83.44	83.49	morph, pos	25.8059 / -
GloVe, DenseNet-CRF	83.96	83.98	morph, character, pos	28.4051 / -
dha DistilBERT(v1), CRF	79.88	82.27	morph, pos	40.2669 / -
dha DistilBERT(v1), LSTM	82.79	83.71	morph, pos	19.8174 / -
dha BERT(v1), BiLSTM-CRF	84.95	85.25	morph, pos	42.1063 / -
dha BERT(v1), BiLSTM	84.51	85.55	morph, pos	18.9292 / -
dha BERT(v1), CRF	82.94	84.99	morph, pos	46.2323 / -
dha BERT(v1)	81.15	84.26	morph, pos	15.1717 / -
dha BERT(v1), BiLSTM-CRF	83.55	83.85	morph, pos	46.0254 / -	del 8,9,10,11
dha BERT(v2), BiLSTM-CRF	83.29	83.57	morph, pos	44.4813 / -
dha-bpe BERT(v1), BiLSTM-CRF	82.83	83.83	morph, pos	42.4347 / -
dha-bpe BERT(v3), BiLSTM-CRF	85.14	85.94	morph, pos	40.1359 / -
dha-bpe BERT(v3), BiLSTM-CRF	85.09	85.90	morph, pos	39.4648 / -	kor-bert-base-dha_bpe.v3.KMOU (fine-tuned)
dha-bpe BERT-large(v1), BiLSTM-CRF	82.86	84.91	morph, pos	53.6760 / -
dha-bpe BERT-large(v3), BiLSTM-CRF	85.87	86.33	morph, pos	50.7508 / -
ELMo, BiLSTM-CRF	85.64	85.66	morph, pos	95.9868 / -
ELMo, BiLSTM-CRF	85.81	85.82	morph, character, pos	95.6196 / -
ELMo, GloVe, BiLSTM-CRF	86.37	86.37	morph, pos	82.7731 / -
ELMo, GloVe, BiLSTM-CRF	86.62	86.63	morph, character, pos	105.739 / -

etagger, measured by conlleval (micro F1)

	m-by-m F1 (%)	e-by-e F1 (%)	Features	Etc
GloVe, BiLSTM-CRF	85.51	85.51	morph, character, pos
dha BERT(v1), BiLSTM-C RF	81.25	81.39	morph, pos	BERT as feature-based
ELMo, GloVe, BiLSTM-CRF	86.75	86.75	morph, character, pos

clova2019_morph_space(morph-based + space as token)

ntagger, measured by conlleval.pl (micro F1)

	m-by-m F1 (%)	e-by-e F1 (%)	Features	GPU / CPU
GloVe, BiLSTM-CRF	85.59	85.72	morph, character, pos	29.0723 / -
dha BERT(v1), BiLSTM-CRF	85.17	85.61	morph, pos	43.7969 / -
ELMo, GloVe, BiLSTM-CRF	85.95	86.06	morph, character, pos	113.177 / -

emb_class=glove, enc_class=bilstm

- train ``` * for clova2019_morph ** token_emb_dim in configs/config-glove.json == 300 (ex, kor.glove.300k.300d.txt ) $ python preprocess.py --data_dir data/clova2019_morph --embedding_path embeddings/kor.glove.300k.300d.txt ** --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding. $ python train.py --save_path=pytorch-model-glove-kor-clova-morph.pt --data_dir data/clova2019_morph --use_crf --embedding_trainable * for clova2019_morph_space $ python preprocess.py --data_dir data/clova2019_morph_space --embedding_path embeddings/kor.glove.300k.300d.txt $ python train.py --save_path=pytorch-model-glove-kor-clova-morph-space.pt --data_dir data/clova2019_morph_space --use_crf --embedding_trainable --use_char_cnn ``` - evaluation ``` * for clova2019_morph $ python evaluate.py --model_path=pytorch-model-glove-kor-clova-morph.pt --data_dir data/clova2019_morph --use_crf ** seqeval.metrics supports IOB2(BIO) format, so FB1 from conlleval.pl should be similar value with. $ cd data/clova2019_morph; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8434531044045398, 9000 INFO:__main__:[Elapsed Time] : 270872ms, 30.096888888888888ms on average accuracy: 93.80%; precision: 84.82%; recall: 83.76%; FB1: 84.29 *** evaluation eoj-by-eoj $ cd data/clova2019_morph ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 93.37%; precision: 84.83%; recall: 83.76%; FB1: 84.29 ** --use_char_cnn INFO:__main__:[F1] : 0.856091088091773, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 225223ms, 25.014668296477385ms on average accuracy: 94.23%; precision: 86.11%; recall: 84.99%; FB1: 85.55 *** evaluation eoj-by-eoj accuracy: 93.88%; precision: 86.11%; recall: 85.00%; FB1: 85.55 ** --warmup_epoch=0 --weight_decay=0.0 --use_char_cnn INFO:__main__:[F1] : 0.8588128417407997, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 233760ms, 25.962329147683075ms on average accuracy: 94.36%; precision: 86.48%; recall: 85.17%; FB1: 85.82 *** evaluation eoj-by-eoj accuracy: 94.01%; precision: 86.48%; recall: 85.17%; FB1: 85.82 * for clova2019_morph_space $ python evaluate.py --model_path=pytorch-model-glove-kor-clova-morph-space.pt --data_dir data/clova2019_morph_space --use_crf --use_char_cnn $ cd data/clova2019_morph_space; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8573393391328268, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 261744ms, 29.072341371263473ms on average accuracy: 95.38%; precision: 86.43%; recall: 84.77%; FB1: 85.59 *** evaluation eoj-by-eoj $ cd data/clova2019_morph_space ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 93.89%; precision: 86.56%; recall: 84.89%; FB1: 85.72 ```

emb_class=glove, enc_class=densenet

- train ``` * token_emb_dim in configs/config-glove.json == 300 (ex, kor.glove.300k.300d.txt ) $ python preprocess.py --config=configs/config-densenet.json --data_dir data/clova2019_morph --embedding_path embeddings/kor.glove.300k.300d.txt * --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding. $ python train.py --config=configs/config-densenet.json --save_path=pytorch-model-densenet-kor-clova-morph.pt --data_dir data/clova2019_morph --use_crf --embedding_trainable ``` - evaluation ``` $ python evaluate.py --config=configs/config-densenet.json --model_path=pytorch-model-densenet-kor-clova-morph.pt --data_dir data/clova2019_morph --use_crf * seqeval.metrics supports IOB2(BIO) format, so FB1 from conlleval.pl should be similar value with. $ cd data/clova2019_morph; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8350127432612621, 9000 INFO:__main__:[Elapsed Time] : 232331ms, 25.805978442049117ms on average accuracy: 93.42%; precision: 82.80%; recall: 84.10%; FB1: 83.44 ** evaluation eoj-by-eoj $ cd data/clova2019_morph ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 92.96%; precision: 82.86%; recall: 84.13%; FB1: 83.49 * --user_char_cnn INFO:__main__:[F1] : 0.8402205267380136, 9000 INFO:__main__:[Elapsed Time] : 255785ms, 28.405156128458717ms on average accuracy: 93.66%; precision: 84.25%; recall: 83.68%; FB1: 83.96 ** evaluation eoj-by-eoj accuracy: 93.24%; precision: 84.28%; recall: 83.69%; FB1: 83.98 ```

emb_class=bert, enc_class=bilstm, eoj-based

- train ``` * n_ctx size should be less than 512 * for clova2019 $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/kor-bert-base-bpe.v1 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/kor-bert-base-bpe.v1 --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/clova2019 --use_crf --eval_steps=1000 ``` - evaluation ``` * for clova2019 $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --use_crf $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. (bpe BERT BiLSTM-CRF) INFO:__main__:[F1] : 0.863551528725101, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 409696.9916820526ms, 45.51278335383924ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4773.615837097168ms, 46.938994918206724ms on average accuracy: 94.33%; precision: 86.17%; recall: 86.52%; FB1: 86.34 ** --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 INFO:__main__:[F1] : 0.872042113153546, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 368655.66992759705ms, 40.952059067226884ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4002.8138160705566ms, 39.17879769296357ms on average accuracy: 94.64%; precision: 87.33%; recall: 87.02%; FB1: 87.17 ** using sub token label, --bert_use_sub_label *** --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 INFO:__main__:[F1] : 0.867761275378117, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 371622.62415885925ms, 41.28106700962074ms on average NFO:__main__:[Elapsed Time] : 100 examples, 3990.586280822754ms, 39.05752692559753ms on average accuracy: 94.63%; precision: 87.15%; recall: 86.84%; FB1: 86.99 ** slicing logits *** --bert_use_crf_slice --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 # https://github.com/dsindex/ntagger/releases/tag/v1.0 INFO:__main__:[F1] : 0.8755258093632384, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 378671.01287841797ms, 42.06319689313522ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4061.053991317749ms, 39.69146506954925ms on average accuracy: 94.73%; precision: 87.50%; recall: 87.39%; FB1: 87.45 ** without --use_crf (bpe BERT BiLSTM) INFO:__main__:[F1] : 0.8646059046587216, 9000 INFO:__main__:[Elapsed Time] : 100 examples, 2224ms, 21.32323232323232ms on average accuracy: 94.31%; precision: 85.92%; recall: 86.82%; FB1: 86.37 ** --bert_disable_lstm (bpe BERT CRF) INFO:__main__:[F1] : 0.8643569376373161, 9000 INFO:__main__:[Elapsed Time] : 342154ms, 38.00722302478053ms on average accuracy: 94.35%; precision: 85.90%; recall: 86.94%; FB1: 86.42 ** --bert_disable_lstm , without --use_crf (bpe BERT) INFO:__main__:[F1] : 0.8677214324767633, 9000 INFO:__main__:[Elapsed Time] : 868094ms, 96.45471719079897ms on average accuracy: 94.47%; precision: 87.02%; recall: 86.33%; FB1: 86.68 ** --bert_disable_lstm , n_ctx=50 INFO:__main__:[F1] : 0.8602524268436113, 9000 INFO:__main__:[Elapsed Time] : 192653ms, 21.39648849872208ms on average accuracy: 93.22%; precision: 85.55%; recall: 83.25%; FB1: 84.38 ** --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --epoch=30 , without --use_crf (bpe BERT) INFO:__main__:[F1] : 0.863227606609181, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 889043ms, 98.78186465162796ms on average accuracy: 94.20%; precision: 85.18%; recall: 87.30%; FB1: 86.23 ** --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 --epoch=30 , without --use_crf (bpe BERT) INFO:__main__:[F1] : 0.8722265771446098, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 952261ms, 105.80508945438382ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1714ms, 16.21212121212121ms on average accuracy: 94.63%; precision: 87.25%; recall: 87.01%; FB1: 87.13 ** --bert_model_name_or_path=./embeddings/kor-bert-large-bpe.v1 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 --epoch=30 , without --use_crf (bpe BERT-large) INFO:__main__:[F1] : 0.8608467232968307, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 1040116.376876831ms, 115.56598331838438ms on average INFO:__main__:[Elapsed Time] : 100 examples, 3212.3892307281494ms, 30.75131984672161ms on average accuracy: 94.13%; precision: 86.19%; recall: 85.79%; FB1: 85.99 ** --bert_model_name_or_path=./embeddings/kor-bert-large-bpe.v1 --warmup_epoch=0 --weight_decay=0.0 --lr=1e-6 --epoch=30 --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 , without --use_crf INFO:__main__:[F1] : 0.8591666808561358, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 984067.9261684418ms, 109.33762731148357ms on average INFO:__main__:[Elapsed Time] : 100 examples, 3314.924716949463ms, 32.00836855955798ms on average accuracy: 94.17%; precision: 85.87%; recall: 85.77%; FB1: 85.82 ** --bert_model_name_or_path=./embeddings/kor-bert-large-bpe.v3 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 --epoch=20 --patience=4 , without --use_crf (bpe BERT-large) INFO:__main__:[F1] : 0.8601462833815275, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 1008693.4926509857ms, 112.0714406355684ms on average INFO:__main__:[Elapsed Time] : 100 examples, 2885.901689529419ms, 27.426498104827573ms on average accuracy: 94.15%; precision: 85.83%; recall: 85.96%; FB1: 85.89 ** --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/kor-distil-bpe-bert.v1 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 --epoch=30 , without --use_crf INFO:__main__:[F1] : 0.852160598843144, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 820104.8340797424ms, 91.12164719606297ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1019.8376178741455ms, 9.317352314188023ms on average accuracy: 93.87%; precision: 85.10%; recall: 85.15%; FB1: 85.12 INFO:__main__:[F1] : 0.85393906493859, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 816257.2820186615ms, 90.68703071211878ms on average INFO:__main__:[Elapsed Time] : 100 examples, 988.6960983276367ms, 9.070220619741113ms on average accuracy: 93.97%; precision: 85.03%; recall: 85.57%; FB1: 85.30 ** --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/kor-distil-wp-bert.v1 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 --epoch=30 , without --use_crf INFO:__main__:[F1] : 0.8458054118245821, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 783889.6675109863ms, 87.09866535082064ms on average INFO:__main__:[Elapsed Time] : 100 examples, 975.9259223937988ms, 8.9646927034012ms on average accuracy: 93.55%; precision: 84.16%; recall: 84.75%; FB1: 84.45 ** --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/distilbert-base-multilingual-cased --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 --epoch=30 , without --use_crf INFO:__main__:[F1] : 0.8407898796667697, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 998458.5220813751ms, 110.94100147581773ms on average accuracy: 93.35%; precision: 84.41%; recall: 83.38%; FB1: 83.89 INFO:__main__:[Elapsed Time] : 100 examples, 1001.7592906951904ms, 9.220564004146691ms on average ** --bert_model_name_or_path=./embeddings/kcbert-base --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 , --without --use_crf (KcBERT-base, BiLSTM) INFO:__main__:[F1] : 0.8491746129396084, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 496907.4342250824ms, 55.207844985460014ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1583.3139419555664ms, 15.055304825908006ms on average accuracy: 93.68%; precision: 85.05%; recall: 84.47%; FB1: 84.76 ** --bert_model_name_or_path=./embeddings/kcbert-base --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 , (KcBERT-base, CRF) INFO:__main__:[F1] : 0.8333361605401095, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 305795.4549789429ms, 33.96846946311906ms on average INFO:__main__:[Elapsed Time] : 100 examples, 3287.224531173706ms, 31.801955868499448ms on average accuracy: 93.10%; precision: 83.13%; recall: 83.51%; FB1: 83.32 ** --bert_model_name_or_path=./embeddings/kcbert-base --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 , --without --use_crf (KcBERT-base) INFO:__main__:[F1] : 0.848736197156657, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 463540.66610336304ms, 51.50017095600344ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1403.444528579712ms, 13.31293462502836ms on average accuracy: 93.75%; precision: 84.90%; recall: 84.54%; FB1: 84.72 ** --bert_model_name_or_path=./embeddings/kcbert-large --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --lr=8e-5 --gradient_accumulation_steps=2 , --without --use_crf (KcBERT-large) INFO:__main__:[F1] : 0.8650133979621444, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 986830.237865448ms, 109.64403990732721ms on average INFO:__main__:[Elapsed Time] : 100 examples, 2811.2776279449463ms, 26.963908262927124ms on average accuracy: 94.31%; precision: 86.53%; recall: 86.16%; FB1: 86.34 ** --bert_model_name_or_path=./embeddings/kor-bert-base-morpheme-aware --bert_disable_lstm --lr=1e-5 --batch_size=32 INFO:__main__:[token classification F1] : 0.8653039030772628, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 839997.5206851959ms, 93.29389646220491ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1754.4901371002197ms, 13.3906759396948ms on average accuracy: 94.34%; precision: 86.23%; recall: 86.57%; FB1: 86.40 ```

emb_class=bert, enc_class=bilstm, morph-based

- train ``` * n_ctx size should be less than 512 * for clova2019_morph ** dha-bpe $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019_morph --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v1 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-clova-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v1 --bert_output_dir=bert-checkpoint-kor-clova-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/clova2019_morph --use_crf --bert_use_pos ** dha $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019_morph --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v2 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-clova-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v2 --bert_output_dir=bert-checkpoint-kor-clova-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/clova2019_morph --use_crf --bert_use_pos $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019_morph --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v1 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-clova-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v1 --bert_output_dir=bert-checkpoint-kor-clova-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/clova2019_morph --use_crf --bert_use_pos * for clova2019_morph_space $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019_morph_space --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v1 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-clova-morph-space.pt --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v1 --bert_output_dir=bert-checkpoint-kor-clova-morph-space --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/clova2019_morph_space --use_crf --bert_use_pos ``` - evaluation ``` * for clova2019_morph ** dha-bpe $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-clova-morph.pt --data_dir=data/clova2019_morph --bert_output_dir=bert-checkpoint-kor-clova-morph --use_crf --bert_use_pos $ cd data/clova2019_morph; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8295019157088124, 9000 INFO:__main__:[Elapsed Time] : 382042ms, 42.434714968329814ms on average accuracy: 93.77%; precision: 81.78%; recall: 83.91%; FB1: 82.83 **** evaluation eoj-by-eoj $ cd data/clova2019_morph ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 93.37%; precision: 83.34%; recall: 84.33%; FB1: 83.83 *** --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v3 --batch_size=64 --warmup_epoch=0 --weight_decay=0.0 --patience=4 INFO:__main__:[F1] : 0.852370233723154, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 361344.5198535919ms, 40.135946203859824ms on average accuracy: 94.60%; precision: 84.67%; recall: 85.61%; FB1: 85.14 **** evaluation eoj-by-eoj accuracy: 94.22%; precision: 85.91%; recall: 85.98%; FB1: 85.94 *** --bert_model_name_or_path=./bert-checkpoint-kor-kmou-morph --batch_size=64 --warmup_epoch=0 --weight_decay=0.0 --patience=4 , fine-tuned kor-bert-base-dha_bpe.v3 on KMOU data INFO:__main__:[F1] : 0.8519647696476965, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 355305.17292022705ms, 39.464807812406406ms on average accuracy: 94.58%; precision: 84.69%; recall: 85.50%; FB1: 85.09 **** evaluation eoj-by-eoj accuracy: 94.21%; precision: 85.95%; recall: 85.86%; FB1: 85.90 *** --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 --warmup_epoch=0 --weight_decay=0.0 --lr=1e-5 INFO:__main__:[F1] : 0.8298886586824331, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 483195.63341140747ms, 53.676061569842936ms on average accuracy: 94.17%; precision: 81.99%; recall: 83.75%; FB1: 82.86 **** evaluation eoj-by-eoj accuracy: 93.85%; precision: 84.97%; recall: 84.85%; FB1: 84.91 *** --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v3 --warmup_epoch=0 --weight_decay=0.0 --lr=5e-5 --patience=4 INFO:__main__:[F1] : 0.8597318852876293, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 456936.72800064087ms, 50.750845153406736ms on average accuracy: 94.66%; precision: 85.63%; recall: 86.11%; FB1: 85.87 **** evaluation eoj-by-eoj accuracy: 94.30%; precision: 86.37%; recall: 86.30%; FB1: 86.33 ** dha $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-clova-morph.pt --data_dir=data/clova2019_morph --bert_output_dir=bert-checkpoint-kor-clova-morph --use_crf --bert_use_pos $ cd data/clova2019_morph; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8336304521299173, 9000 INFO:__main__:[Elapsed Time] : 400446ms, 44.48138682075786ms on average accuracy: 93.58%; precision: 83.12%; recall: 83.46%; FB1: 83.29 **** evaluation eoj-by-eoj $ cd data/clova2019_morph ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 93.06%; precision: 83.55%; recall: 83.59%; FB1: 83.57 $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-clova-morph.pt --data_dir=data/clova2019_morph --bert_output_dir=bert-checkpoint-kor-clova-morph --use_crf --bert_use_pos $ cd data/clova2019_morph; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. (dha BERT BiLSTM-CRF) INFO:__main__:[F1] : 0.850251256281407, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 379036ms, 42.10634514946105ms on average accuracy: 94.30%; precision: 85.06%; recall: 84.84%; FB1: 84.95 *** evaluation eoj-by-eoj $ cd data/clova2019_morph ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 93.86%; precision: 85.50%; recall: 84.99%; FB1: 85.25 ** without --use_crf (dha BERT BiLSTM) INFO:__main__:[F1] : 0.8453688900983027, 9000 INFO:__main__:[Elapsed Time] : 100 examples, 2197ms, 21.11111111111111ms on average accuracy: 94.39%; precision: 84.43%; recall: 84.49%; FB1: 84.46 *** evaluation eoj-by-eoj accuracy: 93.94%; precision: 85.90%; recall: 84.93%; FB1: 85.41 ** --bert_disable_lstm (dha BERT CRF) INFO:__main__:[F1] : 0.8300803673938002, 9000 INFO:__main__:[Elapsed Time] : 100 examples, 4709ms, 46.23232323232323ms on average accuracy: 94.18%; precision: 82.33%; recall: 83.57%; FB1: 82.94 *** evaluation eoj-by-eoj accuracy: 93.84%; precision: 85.15%; recall: 84.83%; FB1: 84.99 ** --bert_disable_lstm , without --use_crf (dha BERT) INFO:__main__:[F1] : 0.8122244286627849, 9000 INFO:__main__:[Elapsed Time] : 100 examples, 1604ms, 15.171717171717171ms on average accuracy: 93.85%; precision: 80.28%; recall: 82.04%; FB1: 81.15 *** evaluation eoj-by-eoj accuracy: 93.48%; precision: 84.92%; recall: 83.61%; FB1: 84.26 ** bert_outputs[2][-7] INFO:__main__:[F1] : 0.8296454550078846, 9000 INFO:__main__:[Elapsed Time] : 376186ms, 41.786642960328926ms on average accuracy: 93.73%; precision: 82.62%; recall: 83.17%; FB1: 82.90 *** evaluation eoj-by-eoj accuracy: 93.19%; precision: 83.25%; recall: 83.34%; FB1: 83.29 ** --bert_remove_layers=8,9,10,11 INFO:__main__:[F1] : 0.8361804271488914, 9000 INFO:__main__:[Elapsed Time] : 414339ms, 46.025447271919106ms on average accuracy: 93.83%; precision: 83.31%; recall: 83.79%; FB1: 83.55 *** evaluation eoj-by-eoj accuracy: 93.31%; precision: 83.78%; recall: 83.93%; FB1: 83.85 ** --warmup_epoch=0 --weight_decay=0.0 --lr=5e-5 --gradient_accumulation_steps=2 --epoch=30 , without --use_crf (dha BERT BiLSTM) INFO:__main__:[F1] : 0.8459056275447281, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 902977ms, 100.3300366707412ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1982ms, 18.92929292929293ms on average accuracy: 94.48%; precision: 83.93%; recall: 85.11%; FB1: 84.51 *** evaluation eoj-by-eoj accuracy: 94.03%; precision: 85.55%; recall: 85.55%; FB1: 85.55 ** --warmup_epoch=0 --weight_decay=0.0 --lr=2e-5 --epoch=30 --bert_disable_lstm , without --use_crf (dha BERT) INFO:__main__:[F1] : 0.8135219179456316, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 887293ms, 98.58706522946994ms on average accuracy: 94.03%; precision: 79.90%; recall: 82.72%; FB1: 81.28 *** evaluation eoj-by-eoj accuracy: 93.69%; precision: 84.50%; recall: 84.63%; FB1: 84.56 ** --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/kor-distil-dha-bert.v1 --warmup_epoch=0 --weight_decay=0.0 --epoch=30 --bert_disable_lstm --lr=8e-5 (1) epoch_0 INFO:__main__:[F1] : 0.7994317797470066, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 331695.18399238586ms, 36.84263667049296ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4152.508020401001ms, 40.266978620278714ms on average accuracy: 93.10%; precision: 79.41%; recall: 80.35%; FB1: 79.88 *** evaluation eoj-by-eoj accuracy: 92.69%; precision: 82.73%; recall: 81.82%; FB1: 82.27 ** --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/kor-distil-dha-bert.v1 --warmup_epoch=0 --weight_decay=0.0 --epoch=30 --lr=8e-5 , without --use_crf (1) epoch_0 INFO:__main__:[F1] : 0.8286240267526895, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 915342.8752422333ms, 101.70127953644766ms on average INFO:__main__:[Elapsed Time] : 100 examples, 2091.184377670288ms, 19.81741250163377ms on average accuracy: 93.82%; precision: 82.19%; recall: 83.40%; FB1: 82.79 *** evaluation eoj-by-eoj accuracy: 93.26%; precision: 83.69%; recall: 83.73%; FB1: 83.71 * for clova2019_morph_space $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-clova-morph-space.pt --data_dir=data/clova2019_morph_space --bert_output_dir=bert-checkpoint-kor-clova-morph-space --use_crf --bert_use_pos $ cd data/clova2019_morph_space; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.853110511030723, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 394271ms, 43.79697744193799ms on average accuracy: 95.51%; precision: 84.96%; recall: 85.38%; FB1: 85.17 *** evaluation eoj-by-eoj $ cd data/clova2019_morph_space ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 93.93%; precision: 85.52%; recall: 85.70%; FB1: 85.61 ```

emb_class=elmo, enc_class=bilstm

- train ``` * for clova2019_morph ** token_emb_dim in configs/config-elmo.json == 300 (ex, kor.glove.300k.300d.txt ) ** elmo_emb_dim in configs/config-elmo.json == 1024 (ex, kor_elmo_2x4096_512_2048cnn_2xhighway_1000k* ) $ python preprocess.py --config=configs/config-elmo.json --data_dir=data/clova2019_morph --embedding_path=embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-elmo.json --save_path=pytorch-model-elmo-kor-clova-morph.pt --data_dir=data/clova2019_morph --elmo_options_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_options.json --elmo_weights_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_weights.hdf5 --use_crf * for clova2019_morph_space $ python preprocess.py --config=configs/config-elmo.json --data_dir=data/clova2019_morph_space --embedding_path=embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-elmo.json --save_path=pytorch-model-elmo-kor-clova-morph-space.pt --data_dir=data/clova2019_morph_space --elmo_options_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_options.json --elmo_weights_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_weights.hdf5 --use_crf --use_char_cnn ``` - evaluation ``` * for clova2019_morph $ python evaluate.py --config=configs/config-elmo.json --model_path=pytorch-model-elmo-kor-clova-morph.pt --data_dir=data/clova2019_morph --elmo_options_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_options.json --elmo_weights_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_weights.hdf5 --use_crf $ cd data/clova2019_morph; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. ** --embedding_trainable INFO:__main__:[F1] : 0.8642865647270933, 9000 INFO:__main__:[Elapsed Time] : 744958ms, 82.7731111111111ms on average accuracy: 94.63%; precision: 86.36%; recall: 86.38%; FB1: 86.37 *** evluation eoj-by-eoj $ cd data/clova2019_morph ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 94.26%; precision: 86.37%; recall: 86.38%; FB1: 86.37 ** --use_char_cnn --embedding_trainable INFO:__main__:[F1] : 0.866860266484455, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 951810ms, 105.7396377375264ms on average accuracy: 94.65%; precision: 86.99%; recall: 86.26%; FB1: 86.62 *** evaluation eoj-by-eoj accuracy: 94.33%; precision: 87.00%; recall: 86.26%; FB1: 86.63 ** modify model.py for disabling glove INFO:__main__:[F1] : 0.856990797967312, 9000 INFO:__main__:[Elapsed Time] : 863968ms, 95.98688743193688ms on average accuracy: 94.30%; precision: 86.44%; recall: 84.85%; FB1: 85.64 *** evaluation eoj-by-eoj accuracy: 93.95%; precision: 86.46%; recall: 84.87%; FB1: 85.66 ** --use_char_cnn , modify model.py for disabling glove INFO:__main__:[F1] : 0.8587286088699316, 9000 INFO:__main__:[Elapsed Time] : 860675ms, 95.61962440271141ms on average accuracy: 94.29%; precision: 86.42%; recall: 85.21%; FB1: 85.81 *** evaluation eoj-by-eoj accuracy: 94.01%; precision: 86.42%; recall: 85.22%; FB1: 85.82 * for clova2019_morph_space $ python evaluate.py --config=configs/config-elmo.json --model_path=pytorch-model-elmo-kor-clova-morph-space.pt --data_dir=data/clova2019_morph_space --elmo_options_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_options.json --elmo_weights_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_weights.hdf5 --use_crf --use_char_cnn $ cd data/clova2019_morph_space; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8609660351595835, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 1018694ms, 113.17768640960107ms on average accuracy: 95.51%; precision: 86.16%; recall: 85.74%; FB1: 85.95 *** evaluation eoj-by-eoj $ cd data/clova2019_morph_space ; python to-eoj.py < test.txt.pred > test.txt.pred.eoj ; perl ../../etc/conlleval.pl < test.txt.pred.eoj ; cd ../.. accuracy: 94.08%; precision: 86.28%; recall: 85.85%; FB1: 86.06 ```

emb_class=electra/roberta, enc_class=bilstm

- train ``` * share config-bert.json if possible * n_ctx size should be less than 512 * for clova2019 ** KoELECTRA-Base $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/koelectra-base-v1-discriminator $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/koelectra-base-v1-discriminator --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/clova2019 --bert_disable_lstm ** bpe ELECTRA-base(v1) $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/kor-electra-base-bpe.v1 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/kor-electra-base-bpe.v1 --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=8e-5 --epoch=30 --data_dir data/clova2019 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --gradient_accumulation_steps=2 ** dhaToken1.large ELECTRA-base # https://github.com/dsindex/ntagger/releases/tag/v1.0 $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/kor-electra-base-dhaToken1.large $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/kor-electra-base-dhaToken1.large --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=8e-5 --epoch=30 --data_dir data/clova2019 --warmup_epoch=0 --weight_decay=0.0 --use_crf --bert_use_crf_slice --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 ** dhaSyllable ELECTRA-base $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/kor-electra-base-dhaSyllable --bert_use_subword_pooling $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/kor-electra-base-dhaSyllable --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=8e-5 --epoch=30 --data_dir data/clova2019 --warmup_epoch=0 --weight_decay=0.0 --use_crf --bert_use_subword_pooling --eval_steps=1000 ** LM-KOR-ELECTRA $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/electra-kor-base $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/electra-kor-base --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=8e-5 --epoch=30 --data_dir data/clova2019 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 --gradient_accumulation_steps=2 ** RoBERTa-base $ python preprocess.py --config=configs/config-roberta.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/kor-roberta-base-bbpe $ python train.py --config=configs/config-roberta.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/kor-roberta-base-bbpe --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=5e-5 --epoch=30 --data_dir data/clova2019 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 ** MA-RoBERTa-base $ python preprocess.py --config=configs/config-roberta.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/kor-roberta-base-morpheme-aware $ python train.py --config=configs/config-roberta.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/kor-roberta-base-morpheme-aware --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=5e-5 --epoch=30 --data_dir data/clova2019 --bert_disable_lstm ** XLM-RoBERTa-base, XLM-RoBERTa-large $ python preprocess.py --config=configs/config-roberta.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/xlm-roberta-base $ python train.py --config=configs/config-roberta.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/xlm-roberta-base --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=5e-5 --epoch=30 --data_dir data/clova2019 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 ** KLUE-RoBERTa-base, KLUE-RoBERTa-large $ python preprocess.py --config=configs/config-roberta.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/klue-roberta-base $ python train.py --config=configs/config-roberta.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/klue-roberta-base --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=5e-5 --epoch=30 --data_dir data/clova2019 --bert_disable_lstm ** Funnel-base $ python preprocess.py --config=configs/config-bert.json --data_dir data/clova2019 --bert_model_name_or_path=./embeddings/funnel-kor-base $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-eoj.pt --bert_model_name_or_path=./embeddings/funnel-kor-base --bert_output_dir=bert-checkpoint-kor-eoj --batch_size=32 --lr=5e-5 --epoch=30 --data_dir data/clova2019 --bert_disable_lstm --warmup_epoch=0 --weight_decay=0.0 ``` - evaluation ``` * for clova2019 ** KoELECTRA-Base $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --bert_disable_lstm $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8601332716652106, 9000 INFO:__main__:[Elapsed Time] : 100 examples, 1667ms, 15.737373737373737ms on average accuracy: 94.17%; precision: 86.46%; recall: 85.36%; FB1: 85.90 *** --warmup_epoch=0 --weight_decay=0.0 --gradient_accumulation_steps=2 --lr=8e-5 , n_ctx=50 INFO:__main__:[F1] : 0.8647250807012538, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 286224ms, 31.79564396044005ms on average accuracy: 93.01%; precision: 85.94%; recall: 82.36%; FB1: 84.12 *** --warmup_epoch=0 --weight_decay=0.0 --gradient_accumulation_steps=2 --lr=8e-5 --epoch=30 INFO:__main__:[F1] : 0.8674485806561278, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 976734ms, 108.52672519168796ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1597ms, 15.16161616161616ms on average accuracy: 94.48%; precision: 86.52%; recall: 86.75%; FB1: 86.64 *** --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --warmup_epoch=0 --weight_decay=0.0 --gradient_accumulation_steps=2 --lr=8e-5 --epoch=30 INFO:__main__:[F1] : 0.8743705005576774, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 878765.2859687805ms, 97.64226169927423ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1558.605432510376ms, 14.81159528096517ms on average accuracy: 94.70%; precision: 86.67%; recall: 87.95%; FB1: 87.31 *** --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --lr=8e-5 --epoch=30 --use_crf --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 , without --bert_disable_lstm INFO:__main__:[F1] : 0.8784531327126347, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 377629.4982433319ms, 41.94557212408231ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4139.477491378784ms, 40.46985838148329ms on average accuracy: 94.84%; precision: 87.78%; recall: 87.73%; FB1: 87.76 *** using sub token label, --bert_use_sub_label **** --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --lr=8e-5 --epoch=30 --use_crf --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 , without --bert_disable_lstm INFO:__main__:[F1] : 0.874318247875669, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 375233.0663204193ms, 41.67938974250991ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4068.805932998657ms, 39.803909532951586ms on average accuracy: 94.70%; precision: 87.71%; recall: 86.93%; FB1: 87.32 *** slicing logits **** --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --lr=8e-5 --epoch=30 --use_crf --bert_use_crf_slice --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 , without --bert_disable_lstm # https://github.com/dsindex/ntagger/releases/tag/v1.0 INFO:__main__:[F1] : 0.8827849438546868, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 398727.7433872223ms, 44.294040061882015ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4092.3564434051514ms, 40.08559265522042ms on average accuracy: 94.92%; precision: 88.19%; recall: 88.07%; FB1: 88.13 *** subword pooling **** --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --lr=8e-5 --epoch=30 --use_crf --bert_use_subword_pooling --eval_steps=1000 , without --bert_disable_lstm INFO:__main__:[F1] : 0.8789345427469033, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 423700.3529071808ms, 47.06374085311135ms on average accuracy: 94.94%; precision: 87.50%; recall: 88.29%; FB1: 87.89 ** LM-KOR-ELECTRA *** --bert_model_name_or_path=./embeddings/electra-kor-base --warmup_epoch=0 --weight_decay=0.0 --gradient_accumulation_steps=2 --lr=8e-5 --epoch=30 INFO:__main__:[F1] : 0.8750042550294449, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 848208.4937095642ms, 94.24447772211201ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1798.86794090271ms, 17.15457800662879ms on average accuracy: 94.75%; precision: 87.38%; recall: 87.39%; FB1: 87.39 *** slicing logits *** --bert_model_name_or_path=./embeddings/electra-kor-base --lr=8e-5 --epoch=30 --use_crf --bert_use_crf_slice --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 , without --bert_disable_lstm # https://github.com/dsindex/ntagger/releases/tag/v1.0 INFO:__main__:[F1] : 0.8761777503683147, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 372020.70689201355ms, 41.32575471608662ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4059.163808822632ms, 39.72473770681054ms on average accuracy: 94.72%; precision: 88.04%; recall: 86.94%; FB1: 87.49 ** bpe ELECTRA-base(v1) $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --bert_disable_lstm $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8657965765827326, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 1025293.7350273132ms, 113.92257348446465ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1881.6356658935547ms, 18.044946169612384ms on average accuracy: 94.43%; precision: 86.76%; recall: 86.16%; FB1: 86.46 ** dhaToken1.large ELECTRA-base # https://github.com/dsindex/ntagger/releases/tag/v1.0 $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --use_crf --bert_use_crf_slice $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8705621049034683, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 407867.568731308ms, 45.30506216694162ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4550.331354141235ms, 44.371419482760956ms on average accuracy: 94.50%; precision: 87.18%; recall: 86.62%; FB1: 86.90 ** dhaSyllable ELECTRA-base $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --use_crf --bert_use_subword_pooling $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8630221340357839, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 369628.701210022ms, 41.056287965584836ms on average accuracy: 94.25%; precision: 86.86%; recall: 85.77%; FB1: 86.31 ** RoBERTa-base $ python evaluate.py --config=configs/config-roberta.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --bert_disable_lstm $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8556474298866252, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 892125.4014968872ms, 99.12149309250948ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1671.3252067565918ms, 15.69863280864677ms on average accuracy: 94.02%; precision: 85.59%; recall: 85.32%; FB1: 85.45 ** MA-RoBERTa-base $ python evaluate.py --config=configs/config-roberta.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --bert_disable_lstm $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[token classification F1] : 0.8793426111810541, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 892001.630783081ms, 99.09754653071838ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1739.4590377807617ms, 15.700003113409485ms on average accuracy: 94.89%; precision: 87.75%; recall: 87.86%; FB1: 87.80 ** XLM-RoBERTa-base, XLM-RoBERTa-large $ python evaluate.py --config=configs/config-roberta.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --bert_disable_lstm $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8699418515423837, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 847419.8191165924ms, 94.15775130147178ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1889.4524574279785ms, 18.13263122481529ms on average accuracy: 94.51%; precision: 86.95%; recall: 86.72%; FB1: 86.84 *** --bert_model_name_or_path=./embeddings/xlm-roberta-large INFO:__main__:[F1] : 0.8716396592474196, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 958549.7736930847ms, 106.50296955191304ms on average INFO:__main__:[Elapsed Time] : 100 examples, 3725.3706455230713ms, 35.952108074920346ms on average accuracy: 94.48%; precision: 86.88%; recall: 87.14%; FB1: 87.01 ** KLUE-RoBERTa-base, KLUE-RoBERTa-large $ python evaluate.py --config=configs/config-roberta.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --bert_disable_lstm $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[token classification F1] : 0.8789022648439093, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 878899.0874290466ms, 97.65359777545304ms on average INFO:__main__:[Elapsed Time] : 100 examples, 1831.2511444091797ms, 17.39305438417377ms on average accuracy: 94.89%; precision: 87.62%; recall: 87.86%; FB1: 87.74 *** --bert_model_name_or_path=./embeddings/klue-roberta-large --lr=1e-5 INFO:__main__:[token classification F1] : 0.8888285044665603, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 1007004.5683383942ms, 111.88273531077292ms on average INFO:__main__:[Elapsed Time] : 100 examples, 3218.8217639923096ms, 30.901489835796934ms on average accuracy: 95.27%; precision: 88.53%; recall: 88.97%; FB1: 88.75 *** subword pooling **** --bert_use_subword_pooling --lr=1e-5 INFO:__main__:[token classification F1] : 0.8783027507377132, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 909895.295381546ms, 101.09652555151798ms on average INFO:__main__:[Elapsed Time] : 100 examples, 2381.690502166748ms, 22.323678238223298ms on average accuracy: 94.94%; precision: 87.62%; recall: 88.04%; FB1: 87.83 **** --bert_use_subword_pooling --bert_model_name_or_path=./embeddings/klue-roberta-large --lr=1e-5 INFO:__main__:[token classification F1] : 0.8874214794873099, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 1073524.353504181ms, 119.2730264486717ms on average INFO:__main__:[Elapsed Time] : 100 examples, 3537.3592376708984ms, 33.802254031402896ms on average accuracy: 95.29%; precision: 88.39%; recall: 89.10%; FB1: 88.74 ** Funnel-base $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-eoj.pt --data_dir data/clova2019 --bert_output_dir=bert-checkpoint-kor-eoj --bert_disable_lstm $ cd data/clova2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8808256677945353, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 1106301.5718460083ms, 122.92021880588051ms on average INFO:__main__:[Elapsed Time] : 100 examples, 4384.497880935669ms, 42.92876551849673ms on average accuracy: 94.98%; precision: 87.87%; recall: 88.06%; FB1: 87.97 *** slicing logits *** --use_crf --bert_use_crf_slice --bert_freezing_epoch=4 --bert_lr_during_freezing=1e-3 --eval_steps=1000 , without --bert_disable_lstm # https://github.com/dsindex/ntagger/releases/tag/v1.0 INFO:__main__:[F1] : 0.8805391055748417, 9000 INFO:__main__:[Elapsed Time] : 9000 examples, 605362.9877567291ms, 67.25383387736339ms on average INFO:__main__:[Elapsed Time] : 100 examples, 8475.262641906738ms, 83.97073697562169ms on average accuracy: 94.80%; precision: 87.89%; recall: 87.96%; FB1: 87.92 ```

KMOU NER (Korean)

KMOU NER 2019

experiments summary

ntagger, measured by conlleval.pl / token_eval.py (micro F1)

	span / token F1 (%)	Features	GPU / CPU	Etc
GloVe, BiLSTM-CRF	85.26 / 86.54	morph, pos	23.0755 / -
GloVe, BiLSTM-CRF	85.93 / 86.41	morph, character, pos	27.7451 / -
GloVe, DenseNet-CRF	85.30 / 86.89	morph, pos	24.0280 / -
GloVe, DenseNet-CRF	85.91 / 86.38	morph, character, pos	22.7710 / -
dha DistilBERT(v1), BiLSTM-CRF	84.20 / 86.94	morph, pos	32.1762 / -
dha DistilBERT(v1), CRF	84.85 / 87.34	morph, pos	27.4700 / -
dha BERT(v1), BiLSTM-CRF	87.56 / 90.47	morph, pos	40.0766 / -
dha BERT(v1), BiLSTM	88.00 / 90.24	morph, pos	23.0388 / -
dha BERT(v1), CRF	88.46 / 90.56	morph, pos	34.1522 / -
dha BERT(v1)	88.04 / 90.64	morph, pos	17.8542 / -
dha BERT(v1), BiLSTM-CRF	83.99 / 87.54	morph, pos	40.5205 / -	del 8,9,10,11
dha BERT(v2), BiLSTM-CRF	85.24 / 87.35	morph, pos	37.7829 / -
dha-bpe BERT(v1), BiLSTM-CRF	85.18 / 88.01	morph, pos	39.0183 / -
dha-bpe BERT(v3), BiLSTM-CRF	88.71 / 91.16	morph, pos	40.5316 / -
dha-bpe BERT(v3), BiLSTM-CRF	88.68 / 90.24	morph, pos	42.6378 / -	subword pooling, word embedding
dha-bpe BERT-large(v1), CRF	89.02 / 91.07	morph, pos	45.1637 / -
dha-bpe BERT-large(v1), CRF	89.14 / 90.85	morph, pos	52.9427 / -	subword pooling, word embedding
dha-bpe BERT-large(v3), CRF	88.62 / 91.17	morph, pos	47.9219 / -
KoELECTRA-Base-v3, CRF	87.12 / 89.93	morph, pos	36.8965 / -
KoELECTRA-Base-v3, BiLSTM-CRF	88.39 / 90.36	morph, pos	42.0473 / -	subword pooling, word embedding
ELMo, BiLSTM-CRF	88.22 / 89.05	morph, pos	128.029 / -
ELMo, BiLSTM-CRF	88.25 / 89.26	morph, character, pos	127.514 / -
ELMo, GloVe, BiLSTM-CRF	88.10 / 88.71	morph, pos	127.989 / -
ELMo, GloVe, BiLSTM-CRF	88.00 / 89.20	morph, character, pos	116.965 / -

etagger, measured by conlleval (micro F1)

	span / token F1 (%)	Features
ELMo, GloVe, BiLSTM-CRF	89.09 / 89.90	morph, character, pos

Pytorch-BERT-CRF-NER, measured by sklearn.metrics (token-level F1)

	token-level macro / micro F1 (%)	Features
KoBERT+CRF	87.56 / 89.70	morph

emb_class=glove, enc_class=bilstm

- train ``` * token_emb_dim in configs/config-glove.json == 300 (ex, kor.glove.300k.300d.txt ) $ python preprocess.py --data_dir data/kmou2019 --embedding_path embeddings/kor.glove.300k.300d.txt * --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding. $ python train.py --save_path=pytorch-model-glove-kor-kmou-morph.pt --data_dir data/kmou2019 --use_crf --embedding_trainable ``` - evaluation ``` $ python evaluate.py --model_path=pytorch-model-glove-kor-kmou-morph.pt --data_dir data/kmou2019 --use_crf * seqeval.metrics supports IOB2(BIO) format, so FB1 from conlleval.pl should be similar value with. $ cd data/kmou2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. * token-level micro F1 $ cd data/kmou2019; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8526331317303437, 927 INFO:__main__:[Elapsed Time] : 927 examples, 21481ms, 23.0755939524838ms on average accuracy: 96.79%; precision: 85.66%; recall: 84.88%; FB1: 85.26 token_eval micro F1: 0.865492518703242 * --use_char_cnn INFO:__main__:[F1] : 0.8593403342700785, 927 INFO:__main__:[Elapsed Time] : 927 examples, 25819ms, 27.7451403887689ms on average accuracy: 96.74%; precision: 86.56%; recall: 85.32%; FB1: 85.93 token_eval micro F1: 0.8641512381845168 * --warmup_epoch=0 --weight_decay=0.0 --use_char_cnn INFO:__main__:[F1] : 0.8589818607372732, 927 INFO:__main__:[Elapsed Time] : 927 examples, 23254ms, 24.571274298056156ms on average accuracy: 96.87%; precision: 85.57%; recall: 86.23%; FB1: 85.90 token_eval micro F1: 0.8703660344962149 ```

emb_class=glove, enc_class=densenet

- train ``` * token_emb_dim in configs/config-glove.json == 300 (ex, kor.glove.300k.300d.txt ) $ python preprocess.py --config=configs/config-densenet.json --data_dir data/kmou2019 --embedding_path embeddings/kor.glove.300k.300d.txt * --use_crf for adding crf layer, --embedding_trainable for fine-tuning pretrained word embedding. $ python train.py --config=configs/config-densenet.json --save_path=pytorch-model-densenet-kor-kmou-morph.pt --data_dir data/kmou2019 --use_crf --embedding_trainable ``` - evaluation ``` $ python evaluate.py --config=configs/config-densenet.json --model_path=pytorch-model-densenet-kor-kmou-morph.pt --data_dir data/kmou2019 --use_crf * seqeval.metrics supports IOB2(BIO) format, so FB1 from conlleval.pl should be similar value with. $ cd data/kmou2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. * token-level micro F1 $ cd data/kmou2019; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8530318602261048, 927 INFO:__main__:[Elapsed Time] : 927 examples, 22368ms, 24.028077753779698ms on average accuracy: 96.85%; precision: 85.29%; recall: 85.32%; FB1: 85.30 token_eval micro F1: 0.8689248895434463 * --use_char_cnn INFO:__main__:[F1] : 0.8591299200947027, 927 INFO:__main__:[Elapsed Time] : 927 examples, 21204ms, 22.771058315334773ms on average accuracy: 96.77%; precision: 86.58%; recall: 85.26%; FB1: 85.91 token_eval micro F1: 0.8638321196460731 ```

emb_class=bert, enc_class=bilstm, morph-based [1]

- train ``` * n_ctx size should be less than 512 * dha (v1) $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2019 --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v1 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v1 --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/kmou2019 --use_crf --bert_use_pos * dha (v2) $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2019 --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v2 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-base-dha.v2 --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/kmou2019 --use_crf --bert_use_pos ``` - evaluation ``` * dha (v1) $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-kmou-morph.pt --data_dir=data/kmou2019 --bert_output_dir=bert-checkpoint-kor-kmou-morph --use_crf --bert_use_pos $ cd data/kmou2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. $ cd data/kmou2019; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. (dha BERT BiLSTM-CRF) INFO:__main__:[F1] : 0.8751814223512336, 927 INFO:__main__:[Elapsed Time] : 927 examples, 37240ms, 40.076673866090715ms on average accuracy: 97.58%; precision: 86.59%; recall: 88.55%; FB1: 87.56 token_eval micro F1: 0.9047362341162879 ** without --use_crf (dha BERT BiLSTM) INFO:__main__:[F1] : 0.8800116635078, 927 INFO:__main__:[Elapsed Time] : 927 examples, 21444ms, 23.03887688984881ms on average accuracy: 97.56%; precision: 87.38%; recall: 88.63%; FB1: 88.00 token_eval micro F1: 0.9024978600887091 ** --bert_disable_lstm (dha BERT CRF) INFO:__main__:[F1] : 0.8844425112367696, 927 INFO:__main__:[Elapsed Time] : 927 examples, 31752ms, 34.152267818574515ms on average accuracy: 97.64%; precision: 87.37%; recall: 89.57%; FB1: 88.46 token_eval micro F1: 0.9056049478160032 ** --warmup_epoch=0 --weight_decay=0.0 --bert_disable_lstm --epoch=30 (dha BERT CRF) INFO:__main__:[F1] : 0.8810802962102512, 927 INFO:__main__:[Elapsed Time] : 927 examples, 30447ms, 32.355291576673864ms on average accuracy: 97.54%; precision: 87.16%; recall: 89.10%; FB1: 88.12 token_eval micro F1: 0.8997748622001397 ** --bert_disable_lstm --lr_deacy_rate=0.9 , without --use_crf (dha BERT) INFO:__main__:[F1] : 0.880439496891716, 927 INFO:__main__:[Elapsed Time] : 927 examples, 16635ms, 17.854211663066955ms on average accuracy: 97.65%; precision: 86.70%; recall: 89.43%; FB1: 88.04 token_eval micro F1: 0.9064737125702409 ** --bert_remove_layers=8,9,10,11 INFO:__main__:[F1] : 0.8392018779342724, 927 INFO:__main__:[Elapsed Time] : 37666ms, 40.52051835853132ms on average accuracy: 96.88%; precision: 83.99%; recall: 83.99%; FB1: 83.99 token_eval micro F1: 0.8754817902934005 ** --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/kor-distil-dha-bert.v1 --warmup_epoch=0 --weight_decay=0.0 --bert_disable_lstm --epoch=30 --lr=8e-5 (1) epoch_0 INFO:__main__:[F1] : 0.8477473562219324, 927 INFO:__main__:[Elapsed Time] : 927 examples, 25547.332048416138ms, 27.470011175579952ms on average accuracy: 96.96%; precision: 83.79%; recall: 85.93%; FB1: 84.85 token_eval micro F1: 0.8734618063617366 ** --config=configs/config-distilbert.json --bert_model_name_or_path=./embeddings/kor-distil-dha-bert.v1 --warmup_epoch=0 --weight_decay=0.0 --epoch=30 --lr=8e-5 (1) epoch_2 INFO:__main__:[F1] : 0.8417381194593809, 927 INFO:__main__:[Elapsed Time] : 927 examples, 29921.59104347229ms, 32.176273945340846ms on average accuracy: 96.85%; precision: 83.36%; recall: 85.05%; FB1: 84.20 token_eval micro F1: 0.8694366635543106 * dha(v2) $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-kmou-morph.pt --data_dir=data/kmou2019 --bert_output_dir=bert-checkpoint-kor-kmou-morph --use_crf --bert_use_pos $ cd data/kmou2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. $ cd data/kmou2019; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8519003931847969, 927 INFO:__main__:[Elapsed Time] : 35123ms, 37.7829373650108ms on average accuracy: 96.83%; precision: 84.59%; recall: 85.90%; FB1: 85.24 token_eval micro F1: 0.8735865242143024 ```

emb_class=bert, enc_class=bilstm, morph-based [2]

- train ``` * n_ctx size should be less than 512 $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2019 --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v1 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v1 --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/kmou2019 --use_crf --bert_use_pos ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-kmou-morph.pt --data_dir=data/kmou2019 --bert_output_dir=bert-checkpoint-kor-kmou-morph --use_crf --bert_use_pos $ cd data/kmou2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. $ cd data/kmou2019; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8517251211861989, 927 INFO:__main__:[Elapsed Time] : 36267ms, 39.018358531317496ms on average accuracy: 97.14%; precision: 82.79%; recall: 87.72%; FB1: 85.18 token_eval micro F1: 0.8801729462631254 * --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v3 --batch_size=64 --warmup_epoch=0 --weight_decay=0.0 --patience=4 INFO:__main__:[F1] : 0.886564996368918, 927 INFO:__main__:[Elapsed Time] : 927 examples, 37696.48361206055ms, 40.53165665455565ms on average accuracy: 97.80%; precision: 87.80%; recall: 89.63%; FB1: 88.71 token_eval micro F1: 0.9116004962779156 * --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 --bert_disable_lstm --lr=1e-5 INFO:__main__:[F1] : 0.8902403706921518, 927 INFO:__main__:[Elapsed Time] : 927 examples, 41991.15180969238ms, 45.16370275880554ms on average accuracy: 97.79%; precision: 87.80%; recall: 90.28%; FB1: 89.02 token_eval micro F1: 0.9107763615295481 * --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v3 --bert_disable_lstm --lr=1e-5 --warmup_epoch=0 --weight_decay=0.0 --patience=4 INFO:__main__:[F1] : 0.8860723030390322, 927 INFO:__main__:[Elapsed Time] : 927 examples, 44546.09036445618ms, 47.92198120389077ms on average accuracy: 97.81%; precision: 86.97%; recall: 90.34%; FB1: 88.62 token_eval micro F1: 0.9117692189657818 * subword pooling, word embedding, --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v3 --batch_size=64 $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2019 --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v3 --bert_use_subword_pooling --bert_use_word_embedding --embedding_path=./embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-base-dha_bpe.v3 --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=64 --lr=5e-5 --epoch=20 --data_dir data/kmou2019 --use_crf --bert_use_pos --bert_use_subword_pooling --bert_use_word_embedding INFO:__main__:[token classification F1] : 0.8868199010765203, 927 INFO:__main__:[Elapsed Time] : 927 examples, 39642.81368255615ms, 42.63782732945024ms on average accuracy: 97.60%; precision: 87.86%; recall: 89.52%; FB1: 88.68 token_eval micro F1: 0.9024785730831597 * subword pooling, word embedding, --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 --bert_disable_lstm --lr=1e-5 $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2019 --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 --bert_use_subword_pooling --bert_use_word_embedding --embedding_path=./embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=1e-5 --epoch=20 --data_dir data/kmou2019 --bert_disable_lstm --use_crf --bert_use_pos --bert_use_subword_pooling --bert_use_word_embedding INFO:__main__:[token classification F1] : 0.8913835836127716, 927 INFO:__main__:[Elapsed Time] : 927 examples, 49218.56093406677ms, 52.94274176686165ms on average accuracy: 97.72%; precision: 88.51%; recall: 89.78%; FB1: 89.14 token_eval micro F1: 0.9085975657027677 ```

emb_class=electra, enc_class=bilstm

- train ``` * n_ctx size should be less than 512 * share config-bert.json $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2019 --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=1e-5 --epoch=20 --data_dir data/kmou2019 --bert_disable_lstm --use_crf --bert_use_pos ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-kmou-morph.pt --data_dir=data/kmou2019 --bert_output_dir=bert-checkpoint-kor-kmou-morph --bert_disable_lstm --use_crf --bert_use_pos $ cd data/kmou2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. $ cd data/kmou2019; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[token classification F1] : 0.868553368768855, 927 INFO:__main__:[Elapsed Time] : 927 examples, 34294.98624801636ms, 36.89658744803773ms on average accuracy: 97.47%; precision: 85.52%; recall: 88.78%; FB1: 87.12 token_eval micro F1: 0.8993278337916285 * subword pooling, word embedding, with lstm $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2019 --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --bert_use_subword_pooling --bert_use_word_embedding --embedding_path=./embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=1e-5 --epoch=20 --data_dir data/kmou2019 --use_crf --bert_use_pos --bert_use_subword_pooling --bert_use_word_embedding INFO:__main__:[token classification F1] : 0.8837545126353791, 927 INFO:__main__:[Elapsed Time] : 927 examples, 39087.4285697937ms, 42.04730853663665ms on average accuracy: 97.63%; precision: 86.96%; recall: 89.87%; FB1: 88.39 token_eval micro F1: 0.9036218538980969 ```

emb_class=elmo, enc_class=bilstm

- train ``` * token_emb_dim in configs/config-elmo.json == 300 (ex, kor.glove.300k.300d.txt ) * elmo_emb_dim in configs/config-elmo.json == 1024 (ex, kor_elmo_2x4096_512_2048cnn_2xhighway_1000k* ) $ python preprocess.py --config=configs/config-elmo.json --data_dir=data/kmou2019 --embedding_path=embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-elmo.json --save_path=pytorch-model-elmo-kor-kmou-morph.pt --data_dir=data/kmou2019 --elmo_options_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_options.json --elmo_weights_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_weights.hdf5 --use_crf ``` - evaluation ``` $ python evaluate.py --config=configs/config-elmo.json --model_path=pytorch-model-elmo-kor-kmou-morph.pt --data_dir=data/kmou2019 --elmo_options_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_options.json --elmo_weights_file=embeddings/kor_elmo_2x4096_512_2048cnn_2xhighway_1000k_weights.hdf5 --use_crf $ cd data/kmou2019; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. $ cd data/kmou2019; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[F1] : 0.8803757522383678, 927 INFO:__main__:[Elapsed Time] : 126180ms, 135.92548596112312ms on average accuracy: 97.30%; precision: 88.00%; recall: 88.08%; FB1: 88.04 token_eval micro F1: 0.8880736809241336 * --batch_size=64 INFO:__main__:[F1] : 0.8809558502112778, 927 INFO:__main__:[Elapsed Time] : 927 examples, 118820ms, 127.98920086393089ms on average accuracy: 97.29%; precision: 87.42%; recall: 88.78%; FB1: 88.10 token_eval micro F1: 0.8871589332511176 * --embedding_trainable INFO:__main__:[F1] : 0.8755125951962508, 927 INFO:__main__:[Elapsed Time] : 125665ms, 135.366090712743ms on average accuracy: 97.33%; precision: 87.32%; recall: 87.78%; FB1: 87.55 token_eval micro F1: 0.8897515527950312 * --use_char_cnn --batch_size=64 INFO:__main__:[F1] : 0.8799648248571009, 927 INFO:__main__:[Elapsed Time] : 927 examples, 108596ms, 116.96544276457884ms on average accuracy: 97.37%; precision: 87.83%; recall: 88.16%; FB1: 88.00 token_eval micro F1: 0.8920632495607668 * --batch_size=64 , modify model.py for disabling glove INFO:__main__:[F1] : 0.8822326125073057, 927 INFO:__main__:[Elapsed Time] : 118884ms, 128.0291576673866ms on average accuracy: 97.35%; precision: 87.79%; recall: 88.66%; FB1: 88.22 token_eval micro F1: 0.8905289052890529 * --use_char_cnn --batch_size=64 , modify model.py for disabling glove INFO:__main__:[F1] : 0.8825072886297376, 927 INFO:__main__:[Elapsed Time] : 118407ms, 127.51403887688984ms on average accuracy: 97.43%; precision: 87.61%; recall: 88.90%; FB1: 88.25 token_eval micro F1: 0.8926606215608773 ```

KMOU NER 2021

experiments summary

ntagger, measured by conlleval.pl / token_eval.py (micro F1)

	span / token F1 (%)	Features	GPU / CPU	Etc
dha-bpe BERT-large(v1), CRF	83.20 / 86.94	morph, pos	55.3803 / -
dha-bpe BERT-large(v1), CRF	83.53 / 87.13	morph, pos	45.5633 / -	subword pooling, word embedding
KoELECTRA-Base-v3, CRF	84.41 / 87.11	morph, pos	38.0129 / -
KoELECTRA-Base-v3, CRF	84.72 / 87.64	morph, pos	40.6136 / -	subword pooling, word embedding
KLUE-RoBERTa-base, CRF	84.39 / 87.78	morph, pos	40.9797 / -	subword pooling, word embedding
KLUE-RoBERTa-large, CRF	85.24 / 88.19	morph, pos	54.2707 / -	subword pooling, word embedding

emb_class=bert, enc_class=bilstm, morph-based

- train ``` * n_ctx size should be less than 512 $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2021 --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/kmou2021 --bert_disable_lstm --use_crf --bert_use_pos ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-kmou-morph.pt --data_dir=data/kmou2021 --bert_output_dir=bert-checkpoint-kor-kmou-morph --bert_disable_lstm --use_crf --bert_use_pos $ cd data/kmou2021; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. $ cd data/kmou2021; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[token classification F1] : 0.8319154700623138, 5299 INFO:__main__:[Elapsed Time] : 5299 examples, 293569.5028305054ms, 55.38036617165289ms on average accuracy: 96.46%; precision: 82.19%; recall: 84.22%; FB1: 83.20 token eval micro F1: 0.8694284374669146 * subword pooling, word embedding $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2021 --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 --bert_use_subword_pooling --bert_use_word_embedding --embedding_path=./embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/kor-bert-large-dha_bpe.v1 --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/kmou2021 --bert_disable_lstm --use_crf --bert_use_pos --bert_use_subword_pooling --bert_use_word_embedding INFO:__main__:[token classification F1] : 0.8351195159671162, 5299 INFO:__main__:[Elapsed Time] : 5299 examples, 241545.16887664795ms, 45.5633999492232ms on average accuracy: 96.52%; precision: 82.65%; recall: 84.43%; FB1: 83.53 token eval micro F1: 0.8713530863449863 ```

emb_class=electra / roberta, enc_class=bilstm

- train ``` * n_ctx size should be less than 512 * share config-bert.json $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2021 --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/kmou2021 --bert_disable_lstm --use_crf --bert_use_pos * KLUE-RoBERTa-base, KLUE-RoBERTa-large $ python preprocess.py --config=configs/config-roberta.json --data_dir data/kmou2021 --bert_model_name_or_path=./embeddings/klue-roberta-base --bert_use_subword_pooling --bert_use_word_embedding --embedding_path=./embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-roberta.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/klue-roberta-base --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=1e-5 --epoch=20 --data_dir data/kmou2021 --bert_disable_lstm --use_crf --bert_use_pos --bert_use_subword_pooling --bert_use_word_embedding ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-kmou-morph.pt --data_dir=data/kmou2021 --bert_output_dir=bert-checkpoint-kor-kmou-morph --bert_disable_lstm --use_crf --bert_use_pos $ cd data/kmou2021; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. $ cd data/kmou2021; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[token classification F1] : 0.8396272154285407, 5299 INFO:__main__:[Elapsed Time] : 5299 examples, 201513.93675804138ms, 38.01290765714087ms on average accuracy: 96.44%; precision: 83.35%; recall: 85.50%; FB1: 84.41 token level micro F1: 0.8711555442477213 * subword pooling, word embedding $ python preprocess.py --config=configs/config-bert.json --data_dir data/kmou2021 --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --bert_use_subword_pooling --bert_use_word_embedding --embedding_path=./embeddings/kor.glove.300k.300d.txt $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-kmou-morph.pt --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --bert_output_dir=bert-checkpoint-kor-kmou-morph --batch_size=32 --lr=5e-5 --epoch=20 --data_dir data/kmou2021 --bert_disable_lstm --use_crf --bert_use_pos --bert_use_subword_pooling --bert_use_word_embedding INFO:__main__:[token classification F1] : 0.8469365663563542, 5299 INFO:__main__:[Elapsed Time] : 5299 examples, 215320.6009864807ms, 40.613625714715ms on average accuracy: 96.64%; precision: 83.63%; recall: 85.84%; FB1: 84.72 token eval micro F1: 0.876451414237003 * KLUE-RoBERTa-base, subword pooling, word embedding INFO:__main__:[token classification F1] : 0.8438759228709738, 5299 INFO:__main__:[Elapsed Time] : 5299 examples, 217291.34964942932ms, 40.979761814522085ms on average accuracy: 96.69%; precision: 83.22%; recall: 85.59%; FB1: 84.39 token level micro F1: 0.8778010784901139 * KLUE-RoBERTa-large, subword pooling, word embedding INFO:__main__:[token classification F1] : 0.8524501593819223, 5299 INFO:__main__:[Elapsed Time] : 5299 examples, 287711.5547657013ms, 54.27072263294996ms on average accuracy: 96.76%; precision: 83.97%; recall: 86.55%; FB1: 85.24 token eval micro F1: 0.8819225186670456 ```

KLUE NER (Korean)

KLUE NER

experiments summary

ntagger, measured by conlleval.pl / token_eval.py (micro F1)

	entity / char F1 (%)	Features	GPU / CPU	Etc
KoELECTRA-Base-v3	83.35 / 89.59	char	33.1255 / -
KLUE-RoBERTa-base	84.20 / 90.06	char	33.3189 / -
KLUE-RoBERTa-large	85.81 / 91.27	char	45.7652 / -
KoELECTRA-Base-v3	88.17 / 92.02	eoj/morph	34.9123 / -	segmented by to-conll.py
KLUE-RoBERTa-base	88.97 / 92.74	eoj/morph	35.0677 / -	segmented by to-conll.py
KLUE-RoBERTa-large	89.69 (macro 88.82) / 93.19	eoj/morph	49.4650 / -	segmented by to-conll.py

emb_class=electra / roberta, enc_class=bilstm

- train ``` * n_ctx size should be less than 512 * share config-bert.json $ python preprocess.py --config=configs/config-bert.json --data_dir data/klue --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator $ python train.py --config=configs/config-bert.json --save_path=pytorch-model-bert-kor-klue.pt --bert_model_name_or_path=./embeddings/koelectra-base-v3-discriminator --bert_output_dir=bert-checkpoint-kor-klue --batch_size=32 --lr=1e-5 --epoch=20 --data_dir data/klue --bert_disable_lstm * KLUE-RoBERTa-base, KLUE-RoBERTa-large $ python preprocess.py --config=configs/config-roberta.json --data_dir data/klue --bert_model_name_or_path=./embeddings/klue-roberta-base $ python train.py --config=configs/config-roberta.json --save_path=pytorch-model-bert-kor-klue.pt --bert_model_name_or_path=./embeddings/klue-roberta-base --bert_output_dir=bert-checkpoint-kor-klue --batch_size=32 --lr=1e-5 --epoch=20 --data_dir data/klue --bert_disable_lstm ``` - evaluation ``` $ python evaluate.py --config=configs/config-bert.json --model_path=pytorch-model-bert-kor-klue.pt --data_dir=data/klue --bert_output_dir=bert-checkpoint-kor-klue --bert_disable_lstm $ cd data/klue; perl ../../etc/conlleval.pl < test.txt.pred ; cd ../.. $ cd data/klue; python ../../etc/token_eval.py < test.txt.pred ; cd ../.. INFO:__main__:[token classification F1] : 0.8334647694153952, 5000 INFO:__main__:[Elapsed Time] : 5000 examples, 165709.34796333313ms, 33.12553837671831ms on average accuracy: 96.92%; precision: 81.53%; recall: 85.25%; FB1: 83.35 char level micro F1: 0.895993426642639 * KLUE-RoBERTa-base INFO:__main__:[token classification F1] : 0.842000137750534, 5000 INFO:__main__:[Elapsed Time] : 5000 examples, 166681.09965324402ms, 33.31893542977089ms on average accuracy: 97.05%; precision: 82.71%; recall: 85.75%; FB1: 84.20 char level micro F1: 0.9006345665357541 * KLUE-RoBERTa-large INFO:__main__:[token classification F1] : 0.8580685186461708, 5000 INFO:__main__:[Elapsed Time] : 5000 examples, 228947.60727882385ms, 45.76522103546381ms on average accuracy: 97.43%; precision: 84.35%; recall: 87.31%; FB1: 85.81 char level micro F1: 0.9127928737180926 * KoELECTRA-Base-v3, segmented INFO:__main__:[token classification F1] : 0.8817226890756302, 5000 INFO:__main__:[Elapsed Time] : 5000 examples, 174636.47651672363ms, 34.912302055748064ms on average accuracy: 97.80%; precision: 87.15%; recall: 89.22%; FB1: 88.17 token level micro F1: 0.9181545590840943 ** evaluation by character-level $ cd data/klue; python to-char.py --file test.txt.pred > test.txt.pred.char; cd ../.. $ cd data/klue; perl ../../etc/conlleval.pl < test.txt.pred.char ; cd ../.. $ cd data/klue; python ../../etc/token_eval.py < test.txt.pred.char ; cd ../.. accuracy: 97.57%; precision: 87.15%; recall: 89.22%; FB1: 88.17 char level micro F1: 0.9202676812160063 * KLUE-RoBERTa-base, segmented INFO:__main__:[token classification F1] : 0.8896767173294957, 5000 INFO:__main__:[Elapsed Time] : 5000 examples, 175425.91190338135ms, 35.06772521496487ms on average accuracy: 97.95%; precision: 88.15%; recall: 89.80%; FB1: 88.97 token level micro F1: 0.9233786340959917 ** evaluation by character-level accuracy: 97.77%; precision: 88.15%; recall: 89.80%; FB1: 88.97 char level micro F1: 0.9274341142020498 * KLUE-RoBERTa-large, segmented micro avg 0.8895 0.9045 0.8969 14113 macro avg 0.8798 0.8969 0.8882 14113 eighted avg 0.8897 0.9045 0.8970 14113 INFO:__main__:[token classification F1] : 0.8969224283305228, 5000 INFO:__main__:[Elapsed Time] : 5000 examples, 247434.663772583ms, 49.46500650761866ms on average accuracy: 98.08%; precision: 88.95%; recall: 90.45%; FB1: 89.69 token level micro F1: 0.9285750945203838 ** evaluation by character-level accuracy: 97.90%; precision: 88.95%; recall: 90.45%; FB1: 89.69 char level micro F1: 0.9319178819301942 ```

Citation

@misc{ntagger,
  author = {dsindex},
  title = {ntagger},
  year = {2020},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/dsindex/ntagger}},
}

dsindex / ntagger

readme

Description

Requirements

Data

CoNLL 2003 (English)

from etagger, CrossWeigh

data/conll2003

data/conll++

data/conll2003_truecase, data/conll++_truecase

Kaggle NER (English)

from entity-annotated-corpus

data/kaggle

GUM (English)

from entity-recognition-datasets

data/gum

Naver NER 2019 (Korean)

from HanBert-NER

data/clova2019

data/clova2019_morph

data/clova2019_morph_space

KMOU NER (Korean)

from KMOU NER

data/kmou2019

data/kmou2021

KLUE NER (Korean)

from KLUE-benchmark

data/klue

Pretrained models

CoNLL 2003 (English)

experiments summary

Kaggle NER (English)

experiments summary

GUM (English)

experiments summary

Naver NER 2019 (Korean)

experiments summary

clova2019(eoj-based)

clova2019_morph(morph-based)

clova2019_morph_space(morph-based + space as token)

KMOU NER (Korean)

KMOU NER 2019

experiments summary

KMOU NER 2021

experiments summary

KLUE NER (Korean)

KLUE NER

experiments summary

Citation