search

Hironsan / anago

Bidirectional LSTM-CRF and ELMo for Named-Entity Recognition, Part-of-Speech Tagging and so on.
https://anago.herokuapp.com/
MIT License
1.48k stars 371 forks source link

what hyperparams/settings for 90.94 f1 on CoNLL 2003 ? #94

Open sravanbabuiitm opened 5 years ago

sravanbabuiitm commented 5 years ago

Hi,

I have tried training NER model using the Anago sequence code. I also used pretrained glove embeddings and followed exact same code as here : https://github.com/Hironsan/anago/blob/master/examples/ner_glove.py

I also tried w2v with word_embedding lower_case=False, but I'm unable to go past 87 F1 score. I tried changing the optimizer from adam to sgd with grad clipping at 5

Can you please let me know the settings used for 90.94 ? I have tried several runs with diff hyperparam settings but still dont go beyond 87.