issues
search
threelittlemonkeys
/
lstm-crf-pytorch
LSTM-CRF in PyTorch
459
stars
101
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Where is the "training data" ?
#24
NianGaozhi
opened
2 years ago
0
Question about rnn-crf and lstm-crf
#23
kkie02
opened
2 years ago
0
train_data
#22
ljh1029
closed
4 years ago
0
how to use this model in my own dataset?
#21
builtfordeve
closed
4 years ago
0
fixed spelling mistakes
#20
ben-bay
closed
4 years ago
0
train error
#19
ben-bay
closed
5 years ago
3
data format (slash in tokens?)
#18
ben-bay
closed
5 years ago
2
the code was running on the GPU.
#17
TianlinZhang668
closed
5 years ago
1
the size of batch_size
#16
TianlinZhang668
closed
5 years ago
2
the means of self.rnn
#15
TianlinZhang668
closed
5 years ago
3
embedding
#14
TianlinZhang668
closed
5 years ago
2
example training data?
#13
ben-bay
closed
5 years ago
1
Some question about the CRF layer
#12
HT-zhang
closed
5 years ago
1
Refactoring
#11
RadagastRotN
closed
5 years ago
0
Question in computing the CRF score in case of partial annotation
#10
nooralahzadeh
closed
4 years ago
0
The code runs very slowly on GPU
#9
lemonhu
closed
5 years ago
1
Different results per run
#8
turangojayev
closed
5 years ago
5
sentece is not padded with SOS, but labels are
#7
abhidipbhattacharyya
closed
6 years ago
1
transition from last word to <END> tag not added in the the score function
#6
zzyxzz
closed
6 years ago
1
fixed a viterbi decode bug
#5
nlpjoe
closed
6 years ago
0
Are there any publicly available training-data datasets?
#4
YANYANYEAH
closed
6 years ago
12
When use lstm_crf, loss is error .. sometimes positive, sometimes negative. sometimes it's a very big number, why
#3
qlwang25
closed
6 years ago
1
for x, y in enumerate(data):
#2
liu-nlper
closed
6 years ago
1
Training_data Format Problem
#1
LeeeeoLiu
closed
6 years ago
5