jiachenwestlake / cross-domain_ner Goto Github PK

View Code? Open in Web Editor NEW

89.0 3.0 26.0 343 KB

Cross-domain NER using cross-domain language modeling, code for ACL 2019 paper

Python 100.00%

transfer-learning cross-domain-ner zero-shot-learning parameter-generation-network language-model domain-adaptation

cross-domain_ner's Introduction

Cross-Domain_NER

Cross-domain NER using cross-domain language modeling, code for ACL 2019 paper.

Introduction

NER is a fundamental task in NLP. Due to the limitation of labeled resources, cross-domain NER has been a challenging task. Most previous work concentrates on the supervised scenario, making use of labeled data for both source and target domains. A disadvantage of such setting is that they can not train for domains which have no labeled data.

We address this issue, using cross-domain LM as a bridge cross-domains for NER domain adaptation. Performing cross-task and cross-domain transfer by designing a novel Parameter Generation Network.

Experiments on CBS SciTech News Dataset show that our model can effectively allow unsupervised domain adaptation, while also can deriving supervised domain adaption between domains with completely different entity types (i.e. news vs. biomedical).

The naive baseline of Single Task Model (STM in paper) mostly followed NCRF++.

For more details, please refer to our paper:

Cross-Domain NER using Cross-Domain Language Modeling
Chen Jia, Xiaobo Liang and Yue Zhang*
(* Corresponding Author)
ACL 2019

Requirements

Python 2 or 3 
PyTorch 0.3

The memory of one GPU should be no less than 8GB to fit the model.

Pretrained Embeddings

GloVe 100-dimension word vectors (Cite from Here).

DataSet

Source-domain:

CoNLL-2003 English NER data.

Target-domain

Unsupervised: CBS SciTech News (test set) (In: `\unsupervised_domain_adaptation\data\news_tech\tech_test).
Supervised: BioNLP13PC dataset and BioNLP13CG dataset.

LM raw data

Retures domain raw data is released together with the CoNLL-2003 dataset.
SciTech news domain raw data Download.
Reuters news domain raw data Download.

Usage

Command

\supervised_domain_adaptation, \unsupervised_domain_adaptationand \combined_SDA_and_UDA can use the following command to make it run.

python main.py --config train.NER.config

The file train.NER.config contains dataset path and model hyperparameters following NCRF++.

Input format

We recommand using the IBOES label style for NER dataset.
We recommand using an input style of one-sentence-per-line for raw data with word segmentation.

Cite:

If you use our data or code, please cite our paper as follows:

@inproceedings{jia2019cross,
  title={Cross-domain ner using cross-domain language modeling},
  author={Jia, Chen and Liang, Xiaobo and Zhang, Yue},
  booktitle={Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics},
  pages={2464--2474},
  year={2019}
  organization={Association for Computational Linguistics}
}

Update

V2. Combining supervised scenario and unsupervised scenario in \combined_SDA_and_UDA.
V1. The previous supervised scenario in \supervised_domain_adaptation;
The previous unsupervised scenario in \unsupervised_domain_adaptation;

cross-domain_ner's People

Contributors

Stargazers

Watchers

cross-domain_ner's Issues

RuntimeError: CUDNN_STATUS_BAD_PARAM

Hello, i got an error like this.

Traceback (most recent call last):
  File "main.py", line 634, in <module>
    train(data)
  File "main.py", line 456, in train
    loss, perplexity, tag_seq_forward, tag_seq_backward, tag_seq = model.loss('model1', batch_word[0], batch_features[0], batch_wordlen[0], batch_char[0], batch_charlen[0], batch_charrecover[0], batch_label[0], lm_seq_tensor[0], mask[0])
  File "/hdisk4t/sym_ll/sym_ll/NER/Cross-Domain_NER-master/supervised_domain_adaptation/model/seqmodel.py", line 60, in loss
    outs_forward, outs_backward, outs = self.word_hidden(mode, word_inputs, feature_inputs, word_seq_lengths, char_inputs, char_seq_lengths, char_seq_recover)
  File "/home/sym_ll/hdisk4t/sym_ll/anaconda3/envs/torch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 357, in __call__
    result = self.forward(*input, **kwargs)
  File "/hdisk4t/sym_ll/sym_ll/NER/Cross-Domain_NER-master/supervised_domain_adaptation/model/wordsequence.py", line 112, in forward
    word_represent = self.wordrep(word_inputs,feature_inputs, word_seq_lengths, char_inputs, char_seq_lengths, char_seq_recover)
  File "/home/sym_ll/hdisk4t/sym_ll/anaconda3/envs/torch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 357, in __call__
    result = self.forward(*input, **kwargs)
  File "/hdisk4t/sym_ll/sym_ll/NER/Cross-Domain_NER-master/supervised_domain_adaptation/model/wordrep.py", line 98, in forward
    char_features = self.char_feature.get_last_hiddens(char_inputs, char_seq_lengths.cpu().numpy())
  File "/hdisk4t/sym_ll/sym_ll/NER/Cross-Domain_NER-master/supervised_domain_adaptation/model/charcnn.py", line 51, in get_last_hiddens
    char_cnn_out = self.char_cnn(char_embeds)
  File "/home/sym_ll/hdisk4t/sym_ll/anaconda3/envs/torch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 357, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/sym_ll/hdisk4t/sym_ll/anaconda3/envs/torch/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 168, in forward
    self.padding, self.dilation, self.groups)
  File "/home/sym_ll/hdisk4t/sym_ll/anaconda3/envs/torch/lib/python3.6/site-packages/torch/nn/functional.py", line 54, in conv1d
    return f(input, weight, bias)
RuntimeError: CUDNN_STATUS_BAD_PARAM

Datasets

Hi jiachen,

Thanks a lot for making your work open-source!

Could you send the preprocessed datasets (BIEOS format) to me? My email is: [email protected].

Thanks,
Zheng

self._backend.RNN

Thank you for opening resource!
But when the code executes to func = self._backend.RNN(...), which is in LSTM_base.py, it comes to "AttributeError: Could not find function class for [RNN]".
And I found that there is not RNN module in self._backend.
How can I solve the problem?
the version of pytorch is 0.3.0, and python is 3.6.

Requirement Versions

Hi - Can you list the required versions for the packages used (torch, scipy, etc)?
Thank you

BIO fromat

Hi,
Is it possible to use the BIO formatted dataset with your code? If so, how can I do that?
Thanks!

in LSTM_base.py, 'LSTM' object has no attribute '_backend', would you mind telling me what's this?

Reproduce the result of UDA

Hi!
I have some trouble when I try to reproduce the result of UDA in paper. Could you help me find where I do it wrong?
Here is my log.
Seed num: 42
status : train
MODEL: train
new vocab size 61057
B-PER
I-PER
E-PER
S-PER
B-LOC
I-LOC
E-LOC
S-LOC
B-ORG
I-ORG
E-ORG
S-ORG
B-MISC
I-MISC
E-MISC
S-MISC
O
B-PER
I-PER
E-PER
S-PER
B-LOC
I-LOC
E-LOC
S-LOC
B-ORG
I-ORG
E-ORG
S-ORG
B-MISC
I-MISC
E-MISC
S-MISC
O
Load pretrained word embedding, dir: ../../Data/pretrain_emb/glove.6B.100d.txt
Embedding:
pretrain word:400000, prefect match:30737, case_match:29809, oov:9482, oov%:0.1354010481372003
Training model...
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
DATA SUMMARY START:
I/O:
MAX SENTENCE LENGTH: 250
MAX WORD LENGTH: -1
Number normalized: True
Word alphabet size: 70029
Char alphabet size: 318
Label alphabet size: 18, Label alphabet size: 18
Word embedding dir: ../../Data/pretrain_emb/glove.6B.100d.txt
Train file directory: data/conll03NER/conll03.train.bmes.space, Train file directory: data/news_tech/tech_test
Dev file directory: data/conll03NER/conll03.dev.bmes.space, Dev file directory: data/news_tech/tech_test
Test file directory: data/conll03NER/conll03.train.bmes.space, Test file directory: data/news_tech/tech_test
Dset file directory: None
Model file directory: data/conll03NER/lstm_crf_ner
Loadmodel directory: data/conll03NER/lstmcrf.model
Decode file directory: None
Train instance number: 14987, Train instance number: 1999
Dev instance number: 3466, Dev instance number: 1999
Test instance number: 14987, Test instance number: 1999
LM instance number: 377591, LM instance number: 398735
++++++++++++++++++++++++++++++++++++++++
Model Network:
Model sl use_crf: True, Model lm use_crf: False
Model word extractor: LSTM
Model use_char: True
Model char extractor: CNN
Model char_hidden_dim: 50
++++++++++++++++++++++++++++++++++++++++
Training:
Optimizer: SGD
Iteration: 100
BatchSize: 10
Average batch loss: True
++++++++++++++++++++++++++++++++++++++++
Hyperparameters:
Hyper lr: 0.015
Hyper lr_decay: 0.05
Hyper HP_clip: None
Hyper momentum: 0.0
Hyper l2: 1e-08
Hyper hidden_dim: 200
Hyper dropout: 0.5
Hyper lstm_layer: 1
Hyper bilstm: True
Hyper GPU: True
DATA SUMMARY END.
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
build network...
build word sequence feature extractor: LSTM...
build word representation...
build char sequence feature extractor: CNN ...
build CRF...
build CRF...
word_hidden.wordrep.char_feature.char_embeddings.weight
word_hidden.wordrep.char_feature.char_cnn.weight
word_hidden.wordrep.char_feature.char_cnn.bias
word_hidden.wordrep.word_embedding.weight
word_hidden.LSTM_param_generator.cpg_.W_task
word_hidden.LSTM_param_generator.task_emb_vocab.weight
word_hidden.LSTM_param_generator.domain_emb_vocab.weight
source_hidden2tag.weight
source_hidden2tag.bias
target_hidden2tag.weight
target_hidden2tag.bias
source_lm_soft_max.softmax_w
source_lm_soft_max.softmax_b
source_crf.transitions
target_crf.transitions
Epoch: 0/100
Learning rate is set as: 0.015
/data3/ywfang/Cross-Domain_NER/unsupervised_domain_adaptation/model/wordsequence.py:90: UserWarning: RNN module weights are not part of single contiguous chunk of memory. This means they need to be compacted at every call, possibly greatly increasing memory usage. To compact weights again call flatten_parameters().
lstm_out, hidden = self.lstm(packed_words, lstm_param, hidden)
/data3/ywfang/Cross-Domain_NER/unsupervised_domain_adaptation/model/wordsequence.py:95: UserWarning: RNN module weights are not part of single contiguous chunk of memory. This means they need to be compacted at every call, possibly greatly increasing memory usage. To compact weights again call flatten_parameters().
lstm_out, hidden = self.lstm(packed_words, lstm_param, hidden)
Epoch: 0 training finished. Time: 2438.25s, speed: 6.15st/s, total loss: 8180.926263153553
Epoch: 0 training finished. Time: 2438.25s, speed: 154.86st/s, total perplexity: 37.2719
Epoch: 0 training finished. Time: 2438.25s, speed: 0.82st/s, total loss: 0
Epoch: 0 training finished. Time: 2438.25s, speed: 163.53st/s, total perplexity: 28.6263
gold_num = 5942 pred_num = 5240 right_num = 4378
gold_num = 2817 pred_num = 1700 right_num = 1097
Dev-Source: time: 11.30s, speed: 483.62st/s; acc: 0.9472, p: 0.8355, r: 0.7368, f: 0.7830
Test-Target: time: 11.30s, speed: 483.62st/s; acc: 0.9503, p: 0.6453, r: 0.3894, f: 0.4857
Exceed previous best f score: -1
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.7830, Target-domain f-score: 0.4857
Epoch: 1/100
Learning rate is set as: 0.014285714285714285
Epoch: 1 training finished. Time: 2464.40s, speed: 6.08st/s, total loss: 4019.5739963948727
Epoch: 1 training finished. Time: 2464.40s, speed: 153.22st/s, total perplexity: 18.5816
Epoch: 1 training finished. Time: 2464.40s, speed: 0.81st/s, total loss: 0
Epoch: 1 training finished. Time: 2464.40s, speed: 161.80st/s, total perplexity: 16.0286
gold_num = 5942 pred_num = 5934 right_num = 5124
gold_num = 2817 pred_num = 3039 right_num = 1860
Dev-Source: time: 11.56s, speed: 472.70st/s; acc: 0.9712, p: 0.8635, r: 0.8623, f: 0.8629
Test-Target: time: 11.56s, speed: 472.70st/s; acc: 0.9616, p: 0.6120, r: 0.6603, f: 0.6352
Exceed previous best f score: 0.7830441781434447
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.8629, Target-domain f-score: 0.6352
Epoch: 2/100
Learning rate is set as: 0.013636363636363634
Epoch: 2 training finished. Time: 2503.11s, speed: 5.99st/s, total loss: 2630.8648856952786
Epoch: 2 training finished. Time: 2503.11s, speed: 150.85st/s, total perplexity: 14.9194
Epoch: 2 training finished. Time: 2503.11s, speed: 0.80st/s, total loss: 0
Epoch: 2 training finished. Time: 2503.11s, speed: 159.30st/s, total perplexity: 13.5252
gold_num = 5942 pred_num = 5927 right_num = 5266
gold_num = 2817 pred_num = 3113 right_num = 1952
Dev-Source: time: 12.31s, speed: 444.09st/s; acc: 0.9761, p: 0.8885, r: 0.8862, f: 0.8874
Test-Target: time: 12.31s, speed: 444.09st/s; acc: 0.9609, p: 0.6270, r: 0.6929, f: 0.6583
Exceed previous best f score: 0.8629168070057258
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.8874, Target-domain f-score: 0.6583
Epoch: 3/100
Learning rate is set as: 0.013043478260869566
Epoch: 3 training finished. Time: 2477.79s, speed: 6.05st/s, total loss: 2181.689256809652
Epoch: 3 training finished. Time: 2477.79s, speed: 152.39st/s, total perplexity: 13.3372
Epoch: 3 training finished. Time: 2477.79s, speed: 0.81st/s, total loss: 0
Epoch: 3 training finished. Time: 2477.79s, speed: 160.92st/s, total perplexity: 12.1384
gold_num = 5942 pred_num = 5920 right_num = 5328
gold_num = 2817 pred_num = 3137 right_num = 1986
Dev-Source: time: 12.55s, speed: 435.60st/s; acc: 0.9775, p: 0.9000, r: 0.8967, f: 0.8983
Test-Target: time: 12.55s, speed: 435.60st/s; acc: 0.9615, p: 0.6331, r: 0.7050, f: 0.6671
Exceed previous best f score: 0.8873536102451766
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.8983, Target-domain f-score: 0.6671
Epoch: 4/100
Learning rate is set as: 0.0125
Epoch: 4 training finished. Time: 2488.92s, speed: 6.02st/s, total loss: 1933.9672123417258
Epoch: 4 training finished. Time: 2488.92s, speed: 151.71st/s, total perplexity: 12.2564
Epoch: 4 training finished. Time: 2488.92s, speed: 0.80st/s, total loss: 0
Epoch: 4 training finished. Time: 2488.92s, speed: 160.20st/s, total perplexity: 11.3815
gold_num = 5942 pred_num = 5947 right_num = 5345
gold_num = 2817 pred_num = 3266 right_num = 2012
Dev-Source: time: 12.26s, speed: 445.77st/s; acc: 0.9783, p: 0.8988, r: 0.8995, f: 0.8992
Test-Target: time: 12.26s, speed: 445.77st/s; acc: 0.9602, p: 0.6160, r: 0.7142, f: 0.6615
Exceed previous best f score: 0.8983308042488619
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.8992, Target-domain f-score: 0.6615
Epoch: 5/100
Learning rate is set as: 0.012
Epoch: 5 training finished. Time: 2935.95s, speed: 5.10st/s, total loss: 1765.3913252688944
Epoch: 5 training finished. Time: 2935.95s, speed: 128.61st/s, total perplexity: 11.3545
Epoch: 5 training finished. Time: 2935.95s, speed: 0.68st/s, total loss: 0
Epoch: 5 training finished. Time: 2935.95s, speed: 135.81st/s, total perplexity: 10.7290
gold_num = 5942 pred_num = 5918 right_num = 5367
gold_num = 2817 pred_num = 3205 right_num = 2023
Dev-Source: time: 15.70s, speed: 348.24st/s; acc: 0.9794, p: 0.9069, r: 0.9032, f: 0.9051
Test-Target: time: 15.70s, speed: 348.24st/s; acc: 0.9612, p: 0.6312, r: 0.7181, f: 0.6719
Exceed previous best f score: 0.8991504752292034
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9051, Target-domain f-score: 0.6719
Epoch: 6/100
Learning rate is set as: 0.011538461538461537
Epoch: 6 training finished. Time: 2844.11s, speed: 5.27st/s, total loss: 1651.9952418655157
Epoch: 6 training finished. Time: 2844.11s, speed: 132.76st/s, total perplexity: 10.9316
Epoch: 6 training finished. Time: 2844.11s, speed: 0.70st/s, total loss: 0
Epoch: 6 training finished. Time: 2844.11s, speed: 140.20st/s, total perplexity: 10.3627
gold_num = 5942 pred_num = 5933 right_num = 5372
gold_num = 2817 pred_num = 3257 right_num = 1986
Dev-Source: time: 14.72s, speed: 371.23st/s; acc: 0.9792, p: 0.9054, r: 0.9041, f: 0.9048
Test-Target: time: 14.72s, speed: 371.23st/s; acc: 0.9608, p: 0.6098, r: 0.7050, f: 0.6539
change optim sgd:
The best Source-domain dev f-score: 0.9051, Target-domain f-score: 0.6719
Epoch: 7/100
Learning rate is set as: 0.01111111111111111
Epoch: 7 training finished. Time: 2831.94s, speed: 5.29st/s, total loss: 1560.4719560779631
Epoch: 7 training finished. Time: 2831.94s, speed: 133.33st/s, total perplexity: 10.4367
Epoch: 7 training finished. Time: 2831.94s, speed: 0.71st/s, total loss: 0
Epoch: 7 training finished. Time: 2831.94s, speed: 140.80st/s, total perplexity: 10.0056
gold_num = 5942 pred_num = 5922 right_num = 5418
gold_num = 2817 pred_num = 3235 right_num = 2056
Dev-Source: time: 14.42s, speed: 378.92st/s; acc: 0.9808, p: 0.9149, r: 0.9118, f: 0.9134
Test-Target: time: 14.42s, speed: 378.92st/s; acc: 0.9610, p: 0.6355, r: 0.7299, f: 0.6794
Exceed previous best f score: 0.9050590219224283
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9134, Target-domain f-score: 0.6794
Epoch: 8/100
Learning rate is set as: 0.010714285714285714
Epoch: 8 training finished. Time: 2841.01s, speed: 5.28st/s, total loss: 1490.763862758875
Epoch: 8 training finished. Time: 2841.01s, speed: 132.91st/s, total perplexity: 10.0947
Epoch: 8 training finished. Time: 2841.01s, speed: 0.70st/s, total loss: 0
Epoch: 8 training finished. Time: 2841.01s, speed: 140.35st/s, total perplexity: 9.7124
gold_num = 5942 pred_num = 5912 right_num = 5427
gold_num = 2817 pred_num = 3187 right_num = 2063
Dev-Source: time: 15.36s, speed: 355.77st/s; acc: 0.9815, p: 0.9180, r: 0.9133, f: 0.9156
Test-Target: time: 15.36s, speed: 355.77st/s; acc: 0.9617, p: 0.6473, r: 0.7323, f: 0.6872
Exceed previous best f score: 0.9133513149022253
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9156, Target-domain f-score: 0.6872
Epoch: 9/100
Learning rate is set as: 0.010344827586206896
Epoch: 9 training finished. Time: 2849.22s, speed: 5.26st/s, total loss: 1418.1109876502305
Epoch: 9 training finished. Time: 2849.22s, speed: 132.52st/s, total perplexity: 9.9065
Epoch: 9 training finished. Time: 2849.22s, speed: 0.70st/s, total loss: 0
Epoch: 9 training finished. Time: 2849.22s, speed: 139.95st/s, total perplexity: 9.4408
gold_num = 5942 pred_num = 5903 right_num = 5435
gold_num = 2817 pred_num = 3171 right_num = 2062
Dev-Source: time: 14.52s, speed: 376.41st/s; acc: 0.9820, p: 0.9207, r: 0.9147, f: 0.9177
Test-Target: time: 14.52s, speed: 376.41st/s; acc: 0.9621, p: 0.6503, r: 0.7320, f: 0.6887
Exceed previous best f score: 0.9156402901974018
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9177, Target-domain f-score: 0.6887
Epoch: 10/100
Learning rate is set as: 0.01
Epoch: 10 training finished. Time: 2838.08s, speed: 5.28st/s, total loss: 1351.7742566540837
Epoch: 10 training finished. Time: 2838.08s, speed: 133.04st/s, total perplexity: 9.6071
Epoch: 10 training finished. Time: 2838.08s, speed: 0.70st/s, total loss: 0
Epoch: 10 training finished. Time: 2838.08s, speed: 140.49st/s, total perplexity: 9.2594
gold_num = 5942 pred_num = 5939 right_num = 5453
gold_num = 2817 pred_num = 3253 right_num = 2041
Dev-Source: time: 17.14s, speed: 318.88st/s; acc: 0.9825, p: 0.9182, r: 0.9177, f: 0.9179
Test-Target: time: 17.14s, speed: 318.88st/s; acc: 0.9606, p: 0.6274, r: 0.7245, f: 0.6725
Exceed previous best f score: 0.917686787674124
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9179, Target-domain f-score: 0.6725
Epoch: 11/100
Learning rate is set as: 0.009677419354838708
Epoch: 11 training finished. Time: 2841.93s, speed: 5.27st/s, total loss: 1289.4590403679758
Epoch: 11 training finished. Time: 2841.93s, speed: 132.86st/s, total perplexity: 9.4231
Epoch: 11 training finished. Time: 2841.93s, speed: 0.70st/s, total loss: 0
Epoch: 11 training finished. Time: 2841.93s, speed: 140.30st/s, total perplexity: 9.0626
gold_num = 5942 pred_num = 5918 right_num = 5457
gold_num = 2817 pred_num = 3225 right_num = 2060
Dev-Source: time: 15.49s, speed: 352.77st/s; acc: 0.9828, p: 0.9221, r: 0.9184, f: 0.9202
Test-Target: time: 15.49s, speed: 352.77st/s; acc: 0.9616, p: 0.6388, r: 0.7313, f: 0.6819
Exceed previous best f score: 0.9179362006565104
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9202, Target-domain f-score: 0.6819
Epoch: 12/100
Learning rate is set as: 0.009375
Epoch: 12 training finished. Time: 2821.85s, speed: 5.31st/s, total loss: 1291.848339015618
Epoch: 12 training finished. Time: 2821.85s, speed: 133.81st/s, total perplexity: 9.2398
Epoch: 12 training finished. Time: 2821.85s, speed: 0.71st/s, total loss: 0
Epoch: 12 training finished. Time: 2821.85s, speed: 141.30st/s, total perplexity: 8.9283
gold_num = 5942 pred_num = 5900 right_num = 5473
gold_num = 2817 pred_num = 3202 right_num = 2070
Dev-Source: time: 14.36s, speed: 380.51st/s; acc: 0.9836, p: 0.9276, r: 0.9211, f: 0.9243
Test-Target: time: 14.36s, speed: 380.51st/s; acc: 0.9626, p: 0.6465, r: 0.7348, f: 0.6878
Exceed previous best f score: 0.9202360876897133
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9243, Target-domain f-score: 0.6878
Epoch: 13/100
Learning rate is set as: 0.00909090909090909
Epoch: 13 training finished. Time: 2830.10s, speed: 5.30st/s, total loss: 1241.1428107935935
Epoch: 13 training finished. Time: 2830.10s, speed: 133.42st/s, total perplexity: 9.1277
Epoch: 13 training finished. Time: 2830.10s, speed: 0.71st/s, total loss: 0
Epoch: 13 training finished. Time: 2830.10s, speed: 140.89st/s, total perplexity: 8.7447
gold_num = 5942 pred_num = 5895 right_num = 5476
gold_num = 2817 pred_num = 3207 right_num = 2070
Dev-Source: time: 14.84s, speed: 368.29st/s; acc: 0.9840, p: 0.9289, r: 0.9216, f: 0.9252
Test-Target: time: 14.84s, speed: 368.29st/s; acc: 0.9623, p: 0.6455, r: 0.7348, f: 0.6873
Exceed previous best f score: 0.924337105218713
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9252, Target-domain f-score: 0.6873
Epoch: 14/100
Learning rate is set as: 0.008823529411764704
Epoch: 14 training finished. Time: 2817.28s, speed: 5.32st/s, total loss: 1195.1746177170426
Epoch: 14 training finished. Time: 2817.28s, speed: 134.03st/s, total perplexity: 8.9837
Epoch: 14 training finished. Time: 2817.28s, speed: 0.71st/s, total loss: 0
Epoch: 14 training finished. Time: 2817.28s, speed: 141.53st/s, total perplexity: 8.6941
gold_num = 5942 pred_num = 5881 right_num = 5478
gold_num = 2817 pred_num = 3156 right_num = 2003
Dev-Source: time: 14.68s, speed: 372.32st/s; acc: 0.9840, p: 0.9315, r: 0.9219, f: 0.9267
Test-Target: time: 14.68s, speed: 372.32st/s; acc: 0.9601, p: 0.6347, r: 0.7110, f: 0.6707
Exceed previous best f score: 0.9252344344006082
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9267, Target-domain f-score: 0.6707
Epoch: 15/100
Learning rate is set as: 0.008571428571428572
Epoch: 15 training finished. Time: 2839.30s, speed: 5.28st/s, total loss: 1161.8517684917897
Epoch: 15 training finished. Time: 2839.30s, speed: 132.99st/s, total perplexity: 8.8109
Epoch: 15 training finished. Time: 2839.30s, speed: 0.70st/s, total loss: 0
Epoch: 15 training finished. Time: 2839.30s, speed: 140.43st/s, total perplexity: 8.5468
gold_num = 5942 pred_num = 5901 right_num = 5502
gold_num = 2817 pred_num = 3204 right_num = 2082
Dev-Source: time: 14.73s, speed: 371.14st/s; acc: 0.9846, p: 0.9324, r: 0.9260, f: 0.9292
Test-Target: time: 14.73s, speed: 371.14st/s; acc: 0.9622, p: 0.6498, r: 0.7391, f: 0.6916
Exceed previous best f score: 0.9266683582846993
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9292, Target-domain f-score: 0.6916
Epoch: 16/100
Learning rate is set as: 0.008333333333333333
Epoch: 16 training finished. Time: 2837.92s, speed: 5.28st/s, total loss: 1137.945263126865
Epoch: 16 training finished. Time: 2837.92s, speed: 133.05st/s, total perplexity: 8.7691
Epoch: 16 training finished. Time: 2837.92s, speed: 0.70st/s, total loss: 0
Epoch: 16 training finished. Time: 2837.92s, speed: 140.50st/s, total perplexity: 8.4371
gold_num = 5942 pred_num = 5920 right_num = 5497
gold_num = 2817 pred_num = 3217 right_num = 2065
Dev-Source: time: 14.11s, speed: 387.50st/s; acc: 0.9841, p: 0.9285, r: 0.9251, f: 0.9268
Test-Target: time: 14.11s, speed: 387.50st/s; acc: 0.9620, p: 0.6419, r: 0.7330, f: 0.6845
change optim sgd:
The best Source-domain dev f-score: 0.9292, Target-domain f-score: 0.6916
Epoch: 17/100
Learning rate is set as: 0.008108108108108107
Epoch: 17 training finished. Time: 2835.51s, speed: 5.29st/s, total loss: 1123.090085344389
Epoch: 17 training finished. Time: 2835.51s, speed: 133.16st/s, total perplexity: 8.5305
Epoch: 17 training finished. Time: 2835.51s, speed: 0.70st/s, total loss: 0
Epoch: 17 training finished. Time: 2835.51s, speed: 140.62st/s, total perplexity: 8.3634
gold_num = 5942 pred_num = 5904 right_num = 5507
gold_num = 2817 pred_num = 3200 right_num = 2054
Dev-Source: time: 14.76s, speed: 370.30st/s; acc: 0.9845, p: 0.9328, r: 0.9268, f: 0.9298
Test-Target: time: 14.76s, speed: 370.30st/s; acc: 0.9614, p: 0.6419, r: 0.7291, f: 0.6827
Exceed previous best f score: 0.9291564637338512
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9298, Target-domain f-score: 0.6827
Epoch: 18/100
Learning rate is set as: 0.007894736842105263
Epoch: 18 training finished. Time: 2803.25s, speed: 5.35st/s, total loss: 1100.1353973187506
Epoch: 18 training finished. Time: 2803.25s, speed: 134.70st/s, total perplexity: 8.5397
Epoch: 18 training finished. Time: 2803.25s, speed: 0.71st/s, total loss: 0
Epoch: 18 training finished. Time: 2803.25s, speed: 142.24st/s, total perplexity: 8.2759
gold_num = 5942 pred_num = 5920 right_num = 5512
gold_num = 2817 pred_num = 3219 right_num = 2082
Dev-Source: time: 14.46s, speed: 378.07st/s; acc: 0.9846, p: 0.9311, r: 0.9276, f: 0.9294
Test-Target: time: 14.46s, speed: 378.07st/s; acc: 0.9623, p: 0.6468, r: 0.7391, f: 0.6899
change optim sgd:
The best Source-domain dev f-score: 0.9298, Target-domain f-score: 0.6827
Epoch: 19/100
Learning rate is set as: 0.007692307692307691
Epoch: 19 training finished. Time: 2792.26s, speed: 5.37st/s, total loss: 1070.8440424762666
Epoch: 19 training finished. Time: 2792.26s, speed: 135.23st/s, total perplexity: 8.4488
Epoch: 19 training finished. Time: 2792.26s, speed: 0.72st/s, total loss: 0
Epoch: 19 training finished. Time: 2792.26s, speed: 142.80st/s, total perplexity: 8.2558
gold_num = 5942 pred_num = 5921 right_num = 5506
gold_num = 2817 pred_num = 3241 right_num = 2048
Dev-Source: time: 15.34s, speed: 356.22st/s; acc: 0.9844, p: 0.9299, r: 0.9266, f: 0.9283
Test-Target: time: 15.34s, speed: 356.22st/s; acc: 0.9609, p: 0.6319, r: 0.7270, f: 0.6761
change optim sgd:
The best Source-domain dev f-score: 0.9298, Target-domain f-score: 0.6827
Epoch: 20/100
Learning rate is set as: 0.0075
Epoch: 20 training finished. Time: 2766.95s, speed: 5.42st/s, total loss: 1050.6705861445516
Epoch: 20 training finished. Time: 2766.95s, speed: 136.46st/s, total perplexity: 8.3147
Epoch: 20 training finished. Time: 2766.95s, speed: 0.72st/s, total loss: 0
Epoch: 20 training finished. Time: 2766.95s, speed: 144.11st/s, total perplexity: 8.1739
gold_num = 5942 pred_num = 5918 right_num = 5519
gold_num = 2817 pred_num = 3209 right_num = 2082
Dev-Source: time: 13.58s, speed: 402.44st/s; acc: 0.9850, p: 0.9326, r: 0.9288, f: 0.9307
Test-Target: time: 13.58s, speed: 402.44st/s; acc: 0.9622, p: 0.6488, r: 0.7391, f: 0.6910
Exceed previous best f score: 0.9297653216275537
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9307, Target-domain f-score: 0.6910
Epoch: 21/100
Learning rate is set as: 0.007317073170731708
Epoch: 21 training finished. Time: 2807.75s, speed: 5.34st/s, total loss: 1056.0537801925093
Epoch: 21 training finished. Time: 2807.75s, speed: 134.48st/s, total perplexity: 8.2599
Epoch: 21 training finished. Time: 2807.75s, speed: 0.71st/s, total loss: 0
Epoch: 21 training finished. Time: 2807.75s, speed: 142.01st/s, total perplexity: 8.0727
gold_num = 5942 pred_num = 5922 right_num = 5525
gold_num = 2817 pred_num = 3243 right_num = 2073
Dev-Source: time: 16.08s, speed: 340.01st/s; acc: 0.9852, p: 0.9330, r: 0.9298, f: 0.9314
Test-Target: time: 16.08s, speed: 340.01st/s; acc: 0.9620, p: 0.6392, r: 0.7359, f: 0.6842
Exceed previous best f score: 0.930691399662732
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9314, Target-domain f-score: 0.6842
Epoch: 22/100
Learning rate is set as: 0.007142857142857143
Epoch: 22 training finished. Time: 2797.93s, speed: 5.36st/s, total loss: 1033.1220356561244
Epoch: 22 training finished. Time: 2797.93s, speed: 134.95st/s, total perplexity: 8.1875
Epoch: 22 training finished. Time: 2797.93s, speed: 0.71st/s, total loss: 0
Epoch: 22 training finished. Time: 2797.93s, speed: 142.51st/s, total perplexity: 8.0291
gold_num = 5942 pred_num = 5920 right_num = 5529
gold_num = 2817 pred_num = 3225 right_num = 2076
Dev-Source: time: 14.29s, speed: 382.42st/s; acc: 0.9851, p: 0.9340, r: 0.9305, f: 0.9322
Test-Target: time: 14.29s, speed: 382.42st/s; acc: 0.9623, p: 0.6437, r: 0.7370, f: 0.6872
Exceed previous best f score: 0.9313890761968981
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9322, Target-domain f-score: 0.6872
Epoch: 23/100
Learning rate is set as: 0.00697674418604651
Epoch: 23 training finished. Time: 2820.35s, speed: 5.31st/s, total loss: 1018.5433005588129
Epoch: 23 training finished. Time: 2820.35s, speed: 133.88st/s, total perplexity: 8.1511
Epoch: 23 training finished. Time: 2820.35s, speed: 0.71st/s, total loss: 0
Epoch: 23 training finished. Time: 2820.35s, speed: 141.38st/s, total perplexity: 7.9828
gold_num = 5942 pred_num = 5907 right_num = 5519
gold_num = 2817 pred_num = 3211 right_num = 2065
Dev-Source: time: 15.07s, speed: 362.73st/s; acc: 0.9851, p: 0.9343, r: 0.9288, f: 0.9316
Test-Target: time: 15.07s, speed: 362.73st/s; acc: 0.9615, p: 0.6431, r: 0.7330, f: 0.6851
change optim sgd:
The best Source-domain dev f-score: 0.9322, Target-domain f-score: 0.6872
Epoch: 24/100
Learning rate is set as: 0.006818181818181817
Epoch: 24 training finished. Time: 2787.56s, speed: 5.38st/s, total loss: 987.0712040197104
Epoch: 24 training finished. Time: 2787.56s, speed: 135.46st/s, total perplexity: 8.0691
Epoch: 24 training finished. Time: 2787.56s, speed: 0.72st/s, total loss: 0
Epoch: 24 training finished. Time: 2787.56s, speed: 143.04st/s, total perplexity: 7.9049
gold_num = 5942 pred_num = 5923 right_num = 5518
gold_num = 2817 pred_num = 3212 right_num = 2061
Dev-Source: time: 13.95s, speed: 391.89st/s; acc: 0.9847, p: 0.9316, r: 0.9286, f: 0.9301
Test-Target: time: 13.95s, speed: 391.89st/s; acc: 0.9620, p: 0.6417, r: 0.7316, f: 0.6837
change optim sgd:
The best Source-domain dev f-score: 0.9322, Target-domain f-score: 0.6872
Epoch: 25/100
Learning rate is set as: 0.006666666666666666
Epoch: 25 training finished. Time: 2816.70s, speed: 5.32st/s, total loss: 981.2934840321541
Epoch: 25 training finished. Time: 2816.70s, speed: 134.05st/s, total perplexity: 8.0302
Epoch: 25 training finished. Time: 2816.70s, speed: 0.71st/s, total loss: 0
Epoch: 25 training finished. Time: 2816.70s, speed: 141.56st/s, total perplexity: 7.8128
gold_num = 5942 pred_num = 5908 right_num = 5528
gold_num = 2817 pred_num = 3196 right_num = 2088
Dev-Source: time: 14.95s, speed: 365.50st/s; acc: 0.9851, p: 0.9357, r: 0.9303, f: 0.9330
Test-Target: time: 14.95s, speed: 365.50st/s; acc: 0.9629, p: 0.6533, r: 0.7412, f: 0.6945
Exceed previous best f score: 0.9322205361659079
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9330, Target-domain f-score: 0.6945
Epoch: 26/100
Learning rate is set as: 0.006521739130434783
Epoch: 26 training finished. Time: 2814.80s, speed: 5.32st/s, total loss: 983.6288790833205
Epoch: 26 training finished. Time: 2814.80s, speed: 134.14st/s, total perplexity: 7.9635
Epoch: 26 training finished. Time: 2814.80s, speed: 0.71st/s, total loss: 0
Epoch: 26 training finished. Time: 2814.80s, speed: 141.66st/s, total perplexity: 7.7903
gold_num = 5942 pred_num = 5908 right_num = 5524
gold_num = 2817 pred_num = 3196 right_num = 2088
Dev-Source: time: 16.36s, speed: 334.09st/s; acc: 0.9850, p: 0.9350, r: 0.9297, f: 0.9323
Test-Target: time: 16.36s, speed: 334.09st/s; acc: 0.9624, p: 0.6533, r: 0.7412, f: 0.6945
change optim sgd:
The best Source-domain dev f-score: 0.9330, Target-domain f-score: 0.6945
Epoch: 27/100
Learning rate is set as: 0.006382978723404255
Epoch: 27 training finished. Time: 2790.06s, speed: 5.37st/s, total loss: 956.2489481503144
Epoch: 27 training finished. Time: 2790.06s, speed: 135.33st/s, total perplexity: 7.9261
Epoch: 27 training finished. Time: 2790.06s, speed: 0.72st/s, total loss: 0
Epoch: 27 training finished. Time: 2790.06s, speed: 142.91st/s, total perplexity: 7.7317
gold_num = 5942 pred_num = 5899 right_num = 5526
gold_num = 2817 pred_num = 3191 right_num = 2072
Dev-Source: time: 14.46s, speed: 378.05st/s; acc: 0.9852, p: 0.9368, r: 0.9300, f: 0.9334
Test-Target: time: 14.46s, speed: 378.05st/s; acc: 0.9618, p: 0.6493, r: 0.7355, f: 0.6897
Exceed previous best f score: 0.9329957805907174
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9334, Target-domain f-score: 0.6897
Epoch: 28/100
Learning rate is set as: 0.006249999999999999
Epoch: 28 training finished. Time: 2808.68s, speed: 5.34st/s, total loss: 936.3749297419563
Epoch: 28 training finished. Time: 2808.68s, speed: 134.44st/s, total perplexity: 7.8348
Epoch: 28 training finished. Time: 2808.68s, speed: 0.71st/s, total loss: 0
Epoch: 28 training finished. Time: 2808.68s, speed: 141.97st/s, total perplexity: 7.7147
gold_num = 5942 pred_num = 5920 right_num = 5526
gold_num = 2817 pred_num = 3212 right_num = 2075
Dev-Source: time: 17.22s, speed: 317.45st/s; acc: 0.9851, p: 0.9334, r: 0.9300, f: 0.9317
Test-Target: time: 17.22s, speed: 317.45st/s; acc: 0.9621, p: 0.6460, r: 0.7366, f: 0.6883
change optim sgd:
The best Source-domain dev f-score: 0.9334, Target-domain f-score: 0.6897
Epoch: 29/100
Learning rate is set as: 0.006122448979591836
Epoch: 29 training finished. Time: 2809.56s, speed: 5.33st/s, total loss: 928.6174706500024
Epoch: 29 training finished. Time: 2809.56s, speed: 134.40st/s, total perplexity: 7.8450
Epoch: 29 training finished. Time: 2809.56s, speed: 0.71st/s, total loss: 0
Epoch: 29 training finished. Time: 2809.56s, speed: 141.92st/s, total perplexity: 7.7393
gold_num = 5942 pred_num = 5929 right_num = 5537
gold_num = 2817 pred_num = 3247 right_num = 2109
Dev-Source: time: 15.44s, speed: 353.96st/s; acc: 0.9852, p: 0.9339, r: 0.9318, f: 0.9329
Test-Target: time: 15.44s, speed: 353.96st/s; acc: 0.9625, p: 0.6495, r: 0.7487, f: 0.6956
change optim sgd:
The best Source-domain dev f-score: 0.9334, Target-domain f-score: 0.6897
Epoch: 30/100
Learning rate is set as: 0.006
Epoch: 30 training finished. Time: 2852.13s, speed: 5.25st/s, total loss: 918.9246400436386
Epoch: 30 training finished. Time: 2852.13s, speed: 132.39st/s, total perplexity: 7.7686
Epoch: 30 training finished. Time: 2852.13s, speed: 0.70st/s, total loss: 0
Epoch: 30 training finished. Time: 2852.13s, speed: 139.80st/s, total perplexity: 7.6891
gold_num = 5942 pred_num = 5922 right_num = 5534
gold_num = 2817 pred_num = 3222 right_num = 2092
Dev-Source: time: 15.45s, speed: 353.76st/s; acc: 0.9854, p: 0.9345, r: 0.9313, f: 0.9329
Test-Target: time: 15.45s, speed: 353.76st/s; acc: 0.9623, p: 0.6493, r: 0.7426, f: 0.6928
change optim sgd:
The best Source-domain dev f-score: 0.9334, Target-domain f-score: 0.6897
Epoch: 31/100
Learning rate is set as: 0.0058823529411764705
Epoch: 31 training finished. Time: 2908.36s, speed: 5.15st/s, total loss: 906.5593660352752
Epoch: 31 training finished. Time: 2908.36s, speed: 129.83st/s, total perplexity: 7.7363
Epoch: 31 training finished. Time: 2908.36s, speed: 0.69st/s, total loss: 0
Epoch: 31 training finished. Time: 2908.36s, speed: 137.10st/s, total perplexity: 7.5846
gold_num = 5942 pred_num = 5924 right_num = 5529
gold_num = 2817 pred_num = 3224 right_num = 2086
Dev-Source: time: 15.11s, speed: 362.22st/s; acc: 0.9850, p: 0.9333, r: 0.9305, f: 0.9319
Test-Target: time: 15.11s, speed: 362.22st/s; acc: 0.9618, p: 0.6470, r: 0.7405, f: 0.6906
change optim sgd:
The best Source-domain dev f-score: 0.9334, Target-domain f-score: 0.6897
Epoch: 32/100
Learning rate is set as: 0.005769230769230769
Epoch: 32 training finished. Time: 2885.52s, speed: 5.19st/s, total loss: 919.9491330655292
Epoch: 32 training finished. Time: 2885.52s, speed: 130.86st/s, total perplexity: 7.6863
Epoch: 32 training finished. Time: 2885.52s, speed: 0.69st/s, total loss: 0
Epoch: 32 training finished. Time: 2885.52s, speed: 138.18st/s, total perplexity: 7.5804
gold_num = 5942 pred_num = 5902 right_num = 5524
gold_num = 2817 pred_num = 3163 right_num = 2091
Dev-Source: time: 15.61s, speed: 350.20st/s; acc: 0.9852, p: 0.9360, r: 0.9297, f: 0.9328
Test-Target: time: 15.61s, speed: 350.20st/s; acc: 0.9631, p: 0.6611, r: 0.7423, f: 0.6993
change optim sgd:
The best Source-domain dev f-score: 0.9334, Target-domain f-score: 0.6897
Epoch: 33/100
Learning rate is set as: 0.005660377358490565
Epoch: 33 training finished. Time: 2836.36s, speed: 5.28st/s, total loss: 891.4518810808659
Epoch: 33 training finished. Time: 2836.36s, speed: 133.13st/s, total perplexity: 7.6718
Epoch: 33 training finished. Time: 2836.36s, speed: 0.70st/s, total loss: 0
Epoch: 33 training finished. Time: 2836.36s, speed: 140.58st/s, total perplexity: 7.5349
gold_num = 5942 pred_num = 5916 right_num = 5533
gold_num = 2817 pred_num = 3200 right_num = 2064
Dev-Source: time: 16.28s, speed: 335.79st/s; acc: 0.9853, p: 0.9353, r: 0.9312, f: 0.9332
Test-Target: time: 16.28s, speed: 335.79st/s; acc: 0.9623, p: 0.6450, r: 0.7327, f: 0.6861
change optim sgd:
The best Source-domain dev f-score: 0.9334, Target-domain f-score: 0.6897
Epoch: 34/100
Learning rate is set as: 0.005555555555555555
Epoch: 34 training finished. Time: 2871.61s, speed: 5.22st/s, total loss: 904.5419793054461
Epoch: 34 training finished. Time: 2871.61s, speed: 131.49st/s, total perplexity: 7.6422
Epoch: 34 training finished. Time: 2871.61s, speed: 0.70st/s, total loss: 0
Epoch: 34 training finished. Time: 2871.61s, speed: 138.85st/s, total perplexity: 7.4879
gold_num = 5942 pred_num = 5909 right_num = 5540
gold_num = 2817 pred_num = 3207 right_num = 2081
Dev-Source: time: 13.90s, speed: 393.10st/s; acc: 0.9858, p: 0.9376, r: 0.9323, f: 0.9349
Test-Target: time: 13.90s, speed: 393.10st/s; acc: 0.9625, p: 0.6489, r: 0.7387, f: 0.6909
Exceed previous best f score: 0.933367114263998
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9349, Target-domain f-score: 0.6909
Epoch: 35/100
Learning rate is set as: 0.005454545454545454
Epoch: 35 training finished. Time: 2873.68s, speed: 5.22st/s, total loss: 865.5962735614739
Epoch: 35 training finished. Time: 2873.68s, speed: 131.40st/s, total perplexity: 7.5529
Epoch: 35 training finished. Time: 2873.68s, speed: 0.70st/s, total loss: 0
Epoch: 35 training finished. Time: 2873.68s, speed: 138.75st/s, total perplexity: 7.5146
gold_num = 5942 pred_num = 5925 right_num = 5552
gold_num = 2817 pred_num = 3218 right_num = 2091
Dev-Source: time: 15.40s, speed: 354.92st/s; acc: 0.9858, p: 0.9370, r: 0.9344, f: 0.9357
Test-Target: time: 15.40s, speed: 354.92st/s; acc: 0.9627, p: 0.6498, r: 0.7423, f: 0.6930
Exceed previous best f score: 0.9349421989705511
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9357, Target-domain f-score: 0.6930
Epoch: 36/100
Learning rate is set as: 0.005357142857142857
Epoch: 36 training finished. Time: 2887.86s, speed: 5.19st/s, total loss: 850.0081989797764
Epoch: 36 training finished. Time: 2887.86s, speed: 130.75st/s, total perplexity: 7.5699
Epoch: 36 training finished. Time: 2887.86s, speed: 0.69st/s, total loss: 0
Epoch: 36 training finished. Time: 2887.86s, speed: 138.07st/s, total perplexity: 7.4746
gold_num = 5942 pred_num = 5914 right_num = 5542
gold_num = 2817 pred_num = 3200 right_num = 2080
Dev-Source: time: 15.35s, speed: 356.09st/s; acc: 0.9857, p: 0.9371, r: 0.9327, f: 0.9349
Test-Target: time: 15.35s, speed: 356.09st/s; acc: 0.9627, p: 0.6500, r: 0.7384, f: 0.6914
change optim sgd:
The best Source-domain dev f-score: 0.9357, Target-domain f-score: 0.6930
Epoch: 37/100
Learning rate is set as: 0.005263157894736842
Epoch: 37 training finished. Time: 2895.59s, speed: 5.18st/s, total loss: 858.0292113730684
Epoch: 37 training finished. Time: 2895.59s, speed: 130.40st/s, total perplexity: 7.5640
Epoch: 37 training finished. Time: 2895.59s, speed: 0.69st/s, total loss: 0
Epoch: 37 training finished. Time: 2895.59s, speed: 137.70st/s, total perplexity: 7.4503
gold_num = 5942 pred_num = 5912 right_num = 5556
gold_num = 2817 pred_num = 3205 right_num = 2103
Dev-Source: time: 14.91s, speed: 366.69st/s; acc: 0.9861, p: 0.9398, r: 0.9350, f: 0.9374
Test-Target: time: 14.91s, speed: 366.69st/s; acc: 0.9632, p: 0.6562, r: 0.7465, f: 0.6984
Exceed previous best f score: 0.9357040532569311
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9374, Target-domain f-score: 0.6984
Epoch: 38/100
Learning rate is set as: 0.005172413793103447
Epoch: 38 training finished. Time: 4369.21s, speed: 3.43st/s, total loss: 847.605830966495
Epoch: 38 training finished. Time: 4369.21s, speed: 86.42st/s, total perplexity: 7.4767
Epoch: 38 training finished. Time: 4369.21s, speed: 0.46st/s, total loss: 0
Epoch: 38 training finished. Time: 4369.21s, speed: 91.26st/s, total perplexity: 7.4749
gold_num = 5942 pred_num = 5927 right_num = 5554
gold_num = 2817 pred_num = 3243 right_num = 2100
Dev-Source: time: 15.01s, speed: 365.28st/s; acc: 0.9859, p: 0.9371, r: 0.9347, f: 0.9359
Test-Target: time: 15.01s, speed: 365.28st/s; acc: 0.9620, p: 0.6475, r: 0.7455, f: 0.6931
change optim sgd:
The best Source-domain dev f-score: 0.9374, Target-domain f-score: 0.6984
Epoch: 39/100
Learning rate is set as: 0.005084745762711864
Epoch: 39 training finished. Time: 2922.17s, speed: 5.13st/s, total loss: 837.7180506261066
Epoch: 39 training finished. Time: 2922.17s, speed: 129.22st/s, total perplexity: 7.4955
Epoch: 39 training finished. Time: 2922.17s, speed: 0.68st/s, total loss: 0
Epoch: 39 training finished. Time: 2922.17s, speed: 136.45st/s, total perplexity: 7.4039
gold_num = 5942 pred_num = 5904 right_num = 5538
gold_num = 2817 pred_num = 3184 right_num = 2059
Dev-Source: time: 16.49s, speed: 332.18st/s; acc: 0.9859, p: 0.9380, r: 0.9320, f: 0.9350
Test-Target: time: 16.49s, speed: 332.18st/s; acc: 0.9619, p: 0.6467, r: 0.7309, f: 0.6862
change optim sgd:
The best Source-domain dev f-score: 0.9374, Target-domain f-score: 0.6984
Epoch: 40/100
Learning rate is set as: 0.005
Epoch: 40 training finished. Time: 2842.98s, speed: 5.27st/s, total loss: 834.5526299457997
Epoch: 40 training finished. Time: 2842.98s, speed: 132.82st/s, total perplexity: 7.4303
Epoch: 40 training finished. Time: 2842.98s, speed: 0.70st/s, total loss: 0
Epoch: 40 training finished. Time: 2842.98s, speed: 140.25st/s, total perplexity: 7.3410
gold_num = 5942 pred_num = 5921 right_num = 5561
gold_num = 2817 pred_num = 3206 right_num = 2092
Dev-Source: time: 14.28s, speed: 382.73st/s; acc: 0.9862, p: 0.9392, r: 0.9359, f: 0.9375
Test-Target: time: 14.28s, speed: 382.73st/s; acc: 0.9623, p: 0.6525, r: 0.7426, f: 0.6947
Exceed previous best f score: 0.9374050953264721
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9375, Target-domain f-score: 0.6947
Epoch: 41/100
Learning rate is set as: 0.0049180327868852455
Epoch: 41 training finished. Time: 2874.82s, speed: 5.21st/s, total loss: 832.0638443576172
Epoch: 41 training finished. Time: 2874.82s, speed: 131.34st/s, total perplexity: 7.4363
Epoch: 41 training finished. Time: 2874.82s, speed: 0.70st/s, total loss: 0
Epoch: 41 training finished. Time: 2874.82s, speed: 138.70st/s, total perplexity: 7.3137
gold_num = 5942 pred_num = 5915 right_num = 5542
gold_num = 2817 pred_num = 3226 right_num = 2083
Dev-Source: time: 16.03s, speed: 340.99st/s; acc: 0.9857, p: 0.9369, r: 0.9327, f: 0.9348
Test-Target: time: 16.03s, speed: 340.99st/s; acc: 0.9621, p: 0.6457, r: 0.7394, f: 0.6894
change optim sgd:
The best Source-domain dev f-score: 0.9375, Target-domain f-score: 0.6947
Epoch: 42/100
Learning rate is set as: 0.004838709677419354
Epoch: 42 training finished. Time: 2851.64s, speed: 5.26st/s, total loss: 821.4504142757505
Epoch: 42 training finished. Time: 2851.64s, speed: 132.41st/s, total perplexity: 7.4249
Epoch: 42 training finished. Time: 2851.64s, speed: 0.70st/s, total loss: 0
Epoch: 42 training finished. Time: 2851.64s, speed: 139.83st/s, total perplexity: 7.3257
gold_num = 5942 pred_num = 5917 right_num = 5557
gold_num = 2817 pred_num = 3222 right_num = 2087
Dev-Source: time: 14.83s, speed: 368.58st/s; acc: 0.9860, p: 0.9392, r: 0.9352, f: 0.9372
Test-Target: time: 14.83s, speed: 368.58st/s; acc: 0.9618, p: 0.6477, r: 0.7409, f: 0.6912
change optim sgd:
The best Source-domain dev f-score: 0.9375, Target-domain f-score: 0.6947
Epoch: 43/100
Learning rate is set as: 0.0047619047619047615
Epoch: 43 training finished. Time: 2916.38s, speed: 5.14st/s, total loss: 810.5953504666686
Epoch: 43 training finished. Time: 2916.38s, speed: 129.47st/s, total perplexity: 7.3126
Epoch: 43 training finished. Time: 2916.38s, speed: 0.69st/s, total loss: 0
Epoch: 43 training finished. Time: 2916.38s, speed: 136.72st/s, total perplexity: 7.2979
gold_num = 5942 pred_num = 5928 right_num = 5550
gold_num = 2817 pred_num = 3236 right_num = 2085
Dev-Source: time: 15.53s, speed: 351.97st/s; acc: 0.9857, p: 0.9362, r: 0.9340, f: 0.9351
Test-Target: time: 15.53s, speed: 351.97st/s; acc: 0.9623, p: 0.6443, r: 0.7401, f: 0.6889
change optim sgd:
The best Source-domain dev f-score: 0.9375, Target-domain f-score: 0.6947
Epoch: 44/100
Learning rate is set as: 0.0046875
Epoch: 44 training finished. Time: 2854.67s, speed: 5.25st/s, total loss: 814.5482726218179
Epoch: 44 training finished. Time: 2854.67s, speed: 132.27st/s, total perplexity: 7.4014
Epoch: 44 training finished. Time: 2854.67s, speed: 0.70st/s, total loss: 0
Epoch: 44 training finished. Time: 2854.67s, speed: 139.68st/s, total perplexity: 7.2942
gold_num = 5942 pred_num = 5917 right_num = 5559
gold_num = 2817 pred_num = 3203 right_num = 2101
Dev-Source: time: 16.00s, speed: 341.51st/s; acc: 0.9862, p: 0.9395, r: 0.9355, f: 0.9375
Test-Target: time: 16.00s, speed: 341.51st/s; acc: 0.9632, p: 0.6559, r: 0.7458, f: 0.6980
change optim sgd:
The best Source-domain dev f-score: 0.9375, Target-domain f-score: 0.6947
Epoch: 45/100
Learning rate is set as: 0.004615384615384615
Epoch: 45 training finished. Time: 2881.27s, speed: 5.20st/s, total loss: 817.0257096197456
Epoch: 45 training finished. Time: 2881.27s, speed: 131.05st/s, total perplexity: 7.3717
Epoch: 45 training finished. Time: 2881.27s, speed: 0.69st/s, total loss: 0
Epoch: 45 training finished. Time: 2881.27s, speed: 138.39st/s, total perplexity: 7.2801
gold_num = 5942 pred_num = 5914 right_num = 5565
gold_num = 2817 pred_num = 3209 right_num = 2106
Dev-Source: time: 14.68s, speed: 372.37st/s; acc: 0.9865, p: 0.9410, r: 0.9366, f: 0.9388
Test-Target: time: 14.68s, speed: 372.37st/s; acc: 0.9630, p: 0.6563, r: 0.7476, f: 0.6990
Exceed previous best f score: 0.9375368793728399
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 46/100
Learning rate is set as: 0.004545454545454545
Epoch: 46 training finished. Time: 2853.91s, speed: 5.25st/s, total loss: 819.4476527716033
Epoch: 46 training finished. Time: 2853.91s, speed: 132.31st/s, total perplexity: 7.3098
Epoch: 46 training finished. Time: 2853.91s, speed: 0.70st/s, total loss: 0
Epoch: 46 training finished. Time: 2853.91s, speed: 139.72st/s, total perplexity: 7.2234
gold_num = 5942 pred_num = 5927 right_num = 5556
gold_num = 2817 pred_num = 3245 right_num = 2083
Dev-Source: time: 15.23s, speed: 358.92st/s; acc: 0.9860, p: 0.9374, r: 0.9350, f: 0.9362
Test-Target: time: 15.23s, speed: 358.92st/s; acc: 0.9614, p: 0.6419, r: 0.7394, f: 0.6872
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 47/100
Learning rate is set as: 0.004477611940298507
Epoch: 47 training finished. Time: 2928.58s, speed: 5.12st/s, total loss: 788.5519095165655
Epoch: 47 training finished. Time: 2928.58s, speed: 128.93st/s, total perplexity: 7.3009
Epoch: 47 training finished. Time: 2928.58s, speed: 0.68st/s, total loss: 0
Epoch: 47 training finished. Time: 2928.58s, speed: 136.15st/s, total perplexity: 7.2319
gold_num = 5942 pred_num = 5922 right_num = 5567
gold_num = 2817 pred_num = 3216 right_num = 2101
Dev-Source: time: 17.71s, speed: 310.03st/s; acc: 0.9864, p: 0.9401, r: 0.9369, f: 0.9385
Test-Target: time: 17.71s, speed: 310.03st/s; acc: 0.9627, p: 0.6533, r: 0.7458, f: 0.6965
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 48/100
Learning rate is set as: 0.004411764705882352
Epoch: 48 training finished. Time: 7034.73s, speed: 2.13st/s, total loss: 792.3497757464647
Epoch: 48 training finished. Time: 7034.73s, speed: 53.68st/s, total perplexity: 7.2680
Epoch: 48 training finished. Time: 7034.73s, speed: 0.28st/s, total loss: 0
Epoch: 48 training finished. Time: 7034.73s, speed: 56.68st/s, total perplexity: 7.1406
gold_num = 5942 pred_num = 5910 right_num = 5559
gold_num = 2817 pred_num = 3198 right_num = 2088
Dev-Source: time: 17.37s, speed: 315.26st/s; acc: 0.9864, p: 0.9406, r: 0.9355, f: 0.9381
Test-Target: time: 17.37s, speed: 315.26st/s; acc: 0.9625, p: 0.6529, r: 0.7412, f: 0.6943
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 49/100
Learning rate is set as: 0.004347826086956521
Epoch: 49 training finished. Time: 2952.83s, speed: 5.08st/s, total loss: 784.3267751713283
Epoch: 49 training finished. Time: 2952.83s, speed: 127.87st/s, total perplexity: 7.2391
Epoch: 49 training finished. Time: 2952.83s, speed: 0.68st/s, total loss: 0
Epoch: 49 training finished. Time: 2952.83s, speed: 135.03st/s, total perplexity: 7.2197
gold_num = 5942 pred_num = 5918 right_num = 5554
gold_num = 2817 pred_num = 3219 right_num = 2108
Dev-Source: time: 14.86s, speed: 367.90st/s; acc: 0.9860, p: 0.9385, r: 0.9347, f: 0.9366
Test-Target: time: 14.86s, speed: 367.90st/s; acc: 0.9626, p: 0.6549, r: 0.7483, f: 0.6985
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 50/100
Learning rate is set as: 0.004285714285714286
Epoch: 50 training finished. Time: 2876.76s, speed: 5.21st/s, total loss: 777.8583422210068
Epoch: 50 training finished. Time: 2876.76s, speed: 131.26st/s, total perplexity: 7.2533
Epoch: 50 training finished. Time: 2876.76s, speed: 0.69st/s, total loss: 0
Epoch: 50 training finished. Time: 2876.76s, speed: 138.61st/s, total perplexity: 7.1517
gold_num = 5942 pred_num = 5915 right_num = 5564
gold_num = 2817 pred_num = 3215 right_num = 2105
Dev-Source: time: 14.66s, speed: 372.95st/s; acc: 0.9864, p: 0.9407, r: 0.9364, f: 0.9385
Test-Target: time: 14.66s, speed: 372.95st/s; acc: 0.9624, p: 0.6547, r: 0.7472, f: 0.6979
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 51/100
Learning rate is set as: 0.004225352112676056
Epoch: 51 training finished. Time: 2884.54s, speed: 5.20st/s, total loss: 766.8275275668129
Epoch: 51 training finished. Time: 2884.54s, speed: 130.90st/s, total perplexity: 7.2199
Epoch: 51 training finished. Time: 2884.54s, speed: 0.69st/s, total loss: 0
Epoch: 51 training finished. Time: 2884.54s, speed: 138.23st/s, total perplexity: 7.1108
gold_num = 5942 pred_num = 5925 right_num = 5561
gold_num = 2817 pred_num = 3232 right_num = 2100
Dev-Source: time: 18.30s, speed: 298.69st/s; acc: 0.9863, p: 0.9386, r: 0.9359, f: 0.9372
Test-Target: time: 18.30s, speed: 298.69st/s; acc: 0.9624, p: 0.6498, r: 0.7455, f: 0.6943
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 52/100
Learning rate is set as: 0.004166666666666667
Epoch: 52 training finished. Time: 2867.37s, speed: 5.23st/s, total loss: 761.6819858364761
Epoch: 52 training finished. Time: 2867.37s, speed: 131.69st/s, total perplexity: 7.2078
Epoch: 52 training finished. Time: 2867.37s, speed: 0.70st/s, total loss: 0
Epoch: 52 training finished. Time: 2867.37s, speed: 139.06st/s, total perplexity: 7.1222
gold_num = 5942 pred_num = 5927 right_num = 5560
gold_num = 2817 pred_num = 3222 right_num = 2101
Dev-Source: time: 15.56s, speed: 351.17st/s; acc: 0.9860, p: 0.9381, r: 0.9357, f: 0.9369
Test-Target: time: 15.56s, speed: 351.17st/s; acc: 0.9628, p: 0.6521, r: 0.7458, f: 0.6958
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 53/100
Learning rate is set as: 0.00410958904109589
Epoch: 53 training finished. Time: 2872.26s, speed: 5.22st/s, total loss: 758.9680653535761
Epoch: 53 training finished. Time: 2872.26s, speed: 131.46st/s, total perplexity: 7.1719
Epoch: 53 training finished. Time: 2872.26s, speed: 0.70st/s, total loss: 0
Epoch: 53 training finished. Time: 2872.26s, speed: 138.82st/s, total perplexity: 7.0981
gold_num = 5942 pred_num = 5921 right_num = 5561
gold_num = 2817 pred_num = 3214 right_num = 2091
Dev-Source: time: 14.60s, speed: 374.29st/s; acc: 0.9863, p: 0.9392, r: 0.9359, f: 0.9375
Test-Target: time: 14.60s, speed: 374.29st/s; acc: 0.9624, p: 0.6506, r: 0.7423, f: 0.6934
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 54/100
Learning rate is set as: 0.004054054054054053
Epoch: 54 training finished. Time: 2867.97s, speed: 5.23st/s, total loss: 763.2362555451691
Epoch: 54 training finished. Time: 2867.97s, speed: 131.66st/s, total perplexity: 7.1165
Epoch: 54 training finished. Time: 2867.97s, speed: 0.70st/s, total loss: 0
Epoch: 54 training finished. Time: 2867.97s, speed: 139.03st/s, total perplexity: 7.0986
gold_num = 5942 pred_num = 5908 right_num = 5554
gold_num = 2817 pred_num = 3179 right_num = 2089
Dev-Source: time: 17.36s, speed: 314.84st/s; acc: 0.9860, p: 0.9401, r: 0.9347, f: 0.9374
Test-Target: time: 17.36s, speed: 314.84st/s; acc: 0.9628, p: 0.6571, r: 0.7416, f: 0.6968
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 55/100
Learning rate is set as: 0.004
Epoch: 55 training finished. Time: 2887.60s, speed: 5.19st/s, total loss: 758.7210296196863
Epoch: 55 training finished. Time: 2887.60s, speed: 130.76st/s, total perplexity: 7.1463
Epoch: 55 training finished. Time: 2887.60s, speed: 0.69st/s, total loss: 0
Epoch: 55 training finished. Time: 2887.60s, speed: 138.09st/s, total perplexity: 7.1285
gold_num = 5942 pred_num = 5926 right_num = 5568
gold_num = 2817 pred_num = 3207 right_num = 2091
Dev-Source: time: 15.29s, speed: 357.51st/s; acc: 0.9864, p: 0.9396, r: 0.9371, f: 0.9383
Test-Target: time: 15.29s, speed: 357.51st/s; acc: 0.9625, p: 0.6520, r: 0.7423, f: 0.6942
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 56/100
Learning rate is set as: 0.003947368421052631
Epoch: 56 training finished. Time: 2885.25s, speed: 5.19st/s, total loss: 752.2264799335971
Epoch: 56 training finished. Time: 2885.25s, speed: 130.87st/s, total perplexity: 7.1006
Epoch: 56 training finished. Time: 2885.25s, speed: 0.69st/s, total loss: 0
Epoch: 56 training finished. Time: 2885.25s, speed: 138.20st/s, total perplexity: 7.0617
gold_num = 5942 pred_num = 5924 right_num = 5563
gold_num = 2817 pred_num = 3197 right_num = 2067
Dev-Source: time: 15.37s, speed: 355.63st/s; acc: 0.9863, p: 0.9391, r: 0.9362, f: 0.9376
Test-Target: time: 15.37s, speed: 355.63st/s; acc: 0.9622, p: 0.6465, r: 0.7338, f: 0.6874
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 57/100
Learning rate is set as: 0.0038961038961038957
Epoch: 57 training finished. Time: 4159.38s, speed: 3.60st/s, total loss: 748.7329716961831
Epoch: 57 training finished. Time: 4159.38s, speed: 90.78st/s, total perplexity: 7.0826
Epoch: 57 training finished. Time: 4159.38s, speed: 0.48st/s, total loss: 0
Epoch: 57 training finished. Time: 4159.38s, speed: 95.86st/s, total perplexity: 7.0230
gold_num = 5942 pred_num = 5912 right_num = 5560
gold_num = 2817 pred_num = 3184 right_num = 2058
Dev-Source: time: 14.80s, speed: 369.95st/s; acc: 0.9863, p: 0.9405, r: 0.9357, f: 0.9381
Test-Target: time: 14.80s, speed: 369.95st/s; acc: 0.9618, p: 0.6464, r: 0.7306, f: 0.6859
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 58/100
Learning rate is set as: 0.0038461538461538455
Epoch: 58 training finished. Time: 2838.79s, speed: 5.28st/s, total loss: 733.6096067801118
Epoch: 58 training finished. Time: 2838.79s, speed: 133.01st/s, total perplexity: 7.0504
Epoch: 58 training finished. Time: 2838.79s, speed: 0.70st/s, total loss: 0
Epoch: 58 training finished. Time: 2838.79s, speed: 140.46st/s, total perplexity: 7.0547
gold_num = 5942 pred_num = 5921 right_num = 5566
gold_num = 2817 pred_num = 3205 right_num = 2086
Dev-Source: time: 16.10s, speed: 339.43st/s; acc: 0.9864, p: 0.9400, r: 0.9367, f: 0.9384
Test-Target: time: 16.10s, speed: 339.43st/s; acc: 0.9626, p: 0.6509, r: 0.7405, f: 0.6928
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 59/100
Learning rate is set as: 0.0037974683544303796
Epoch: 59 training finished. Time: 2848.66s, speed: 5.26st/s, total loss: 728.4975908724591
Epoch: 59 training finished. Time: 2848.66s, speed: 132.55st/s, total perplexity: 7.0584
Epoch: 59 training finished. Time: 2848.66s, speed: 0.70st/s, total loss: 0
Epoch: 59 training finished. Time: 2848.66s, speed: 139.97st/s, total perplexity: 7.0597
gold_num = 5942 pred_num = 5913 right_num = 5559
gold_num = 2817 pred_num = 3209 right_num = 2067
Dev-Source: time: 15.04s, speed: 363.30st/s; acc: 0.9863, p: 0.9401, r: 0.9355, f: 0.9378
Test-Target: time: 15.04s, speed: 363.30st/s; acc: 0.9618, p: 0.6441, r: 0.7338, f: 0.6860
change optim sgd:
The best Source-domain dev f-score: 0.9388, Target-domain f-score: 0.6990
Epoch: 60/100
Learning rate is set as: 0.00375
Epoch: 60 training finished. Time: 2857.92s, speed: 5.24st/s, total loss: 731.9911797801033
Epoch: 60 training finished. Time: 2857.92s, speed: 132.12st/s, total perplexity: 7.0557
Epoch: 60 training finished. Time: 2857.92s, speed: 0.70st/s, total loss: 0
Epoch: 60 training finished. Time: 2857.92s, speed: 139.52st/s, total perplexity: 6.9944
gold_num = 5942 pred_num = 5917 right_num = 5568
gold_num = 2817 pred_num = 3207 right_num = 2111
Dev-Source: time: 15.40s, speed: 354.90st/s; acc: 0.9866, p: 0.9410, r: 0.9371, f: 0.9390
Test-Target: time: 15.40s, speed: 354.90st/s; acc: 0.9630, p: 0.6582, r: 0.7494, f: 0.7009
Exceed previous best f score: 0.9387651821862348
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9390, Target-domain f-score: 0.7009
Epoch: 61/100
Learning rate is set as: 0.003703703703703703
Epoch: 61 training finished. Time: 2858.73s, speed: 5.24st/s, total loss: 731.1324182171375
Epoch: 61 training finished. Time: 2858.73s, speed: 132.08st/s, total perplexity: 7.0040
Epoch: 61 training finished. Time: 2858.73s, speed: 0.70st/s, total loss: 0
Epoch: 61 training finished. Time: 2858.73s, speed: 139.48st/s, total perplexity: 7.0417
gold_num = 5942 pred_num = 5923 right_num = 5568
gold_num = 2817 pred_num = 3220 right_num = 2085
Dev-Source: time: 14.46s, speed: 378.07st/s; acc: 0.9865, p: 0.9401, r: 0.9371, f: 0.9386
Test-Target: time: 14.46s, speed: 378.07st/s; acc: 0.9618, p: 0.6475, r: 0.7401, f: 0.6907
change optim sgd:
The best Source-domain dev f-score: 0.9390, Target-domain f-score: 0.7009
Epoch: 62/100
Learning rate is set as: 0.003658536585365854
Epoch: 62 training finished. Time: 2867.81s, speed: 5.23st/s, total loss: 703.4135433081537
Epoch: 62 training finished. Time: 2867.81s, speed: 131.67st/s, total perplexity: 7.0249
Epoch: 62 training finished. Time: 2867.81s, speed: 0.70st/s, total loss: 0
Epoch: 62 training finished. Time: 2867.81s, speed: 139.04st/s, total perplexity: 6.9692
gold_num = 5942 pred_num = 5921 right_num = 5566
gold_num = 2817 pred_num = 3212 right_num = 2091
Dev-Source: time: 15.57s, speed: 350.98st/s; acc: 0.9864, p: 0.9400, r: 0.9367, f: 0.9384
Test-Target: time: 15.57s, speed: 350.98st/s; acc: 0.9620, p: 0.6510, r: 0.7423, f: 0.6936
change optim sgd:
The best Source-domain dev f-score: 0.9390, Target-domain f-score: 0.7009
Epoch: 63/100
Learning rate is set as: 0.003614457831325301
Epoch: 63 training finished. Time: 2864.00s, speed: 5.23st/s, total loss: 724.4953144593164
Epoch: 63 training finished. Time: 2864.00s, speed: 131.84st/s, total perplexity: 7.0321
Epoch: 63 training finished. Time: 2864.00s, speed: 0.70st/s, total loss: 0
Epoch: 63 training finished. Time: 2864.00s, speed: 139.22st/s, total perplexity: 7.0042
gold_num = 5942 pred_num = 5927 right_num = 5563
gold_num = 2817 pred_num = 3220 right_num = 2063
Dev-Source: time: 18.11s, speed: 301.87st/s; acc: 0.9864, p: 0.9386, r: 0.9362, f: 0.9374
Test-Target: time: 18.11s, speed: 301.87st/s; acc: 0.9607, p: 0.6407, r: 0.7323, f: 0.6835
change optim sgd:
The best Source-domain dev f-score: 0.9390, Target-domain f-score: 0.7009
Epoch: 64/100
Learning rate is set as: 0.0035714285714285713
Epoch: 64 training finished. Time: 4679.79s, speed: 3.20st/s, total loss: 712.5771029810421
Epoch: 64 training finished. Time: 4679.79s, speed: 80.69st/s, total perplexity: 7.0191
Epoch: 64 training finished. Time: 4679.79s, speed: 0.43st/s, total loss: 0
Epoch: 64 training finished. Time: 4679.79s, speed: 85.20st/s, total perplexity: 6.9322
gold_num = 5942 pred_num = 5918 right_num = 5567
gold_num = 2817 pred_num = 3210 right_num = 2079
Dev-Source: time: 14.91s, speed: 366.99st/s; acc: 0.9865, p: 0.9407, r: 0.9369, f: 0.9388
Test-Target: time: 14.91s, speed: 366.99st/s; acc: 0.9619, p: 0.6477, r: 0.7380, f: 0.6899
change optim sgd:
The best Source-domain dev f-score: 0.9390, Target-domain f-score: 0.7009
Epoch: 65/100
Learning rate is set as: 0.003529411764705882
Epoch: 65 training finished. Time: 2835.26s, speed: 5.29st/s, total loss: 700.9482047494967
Epoch: 65 training finished. Time: 2835.26s, speed: 133.18st/s, total perplexity: 6.9756
Epoch: 65 training finished. Time: 2835.26s, speed: 0.71st/s, total loss: 0
Epoch: 65 training finished. Time: 2835.26s, speed: 140.63st/s, total perplexity: 6.9555
gold_num = 5942 pred_num = 5907 right_num = 5562
gold_num = 2817 pred_num = 3195 right_num = 2106
Dev-Source: time: 14.63s, speed: 373.53st/s; acc: 0.9864, p: 0.9416, r: 0.9360, f: 0.9388
Test-Target: time: 14.63s, speed: 373.53st/s; acc: 0.9632, p: 0.6592, r: 0.7476, f: 0.7006
change optim sgd:
The best Source-domain dev f-score: 0.9390, Target-domain f-score: 0.7009
Epoch: 66/100
Learning rate is set as: 0.003488372093023255
Epoch: 66 training finished. Time: 2883.78s, speed: 5.20st/s, total loss: 710.3381256414577
Epoch: 66 training finished. Time: 2883.78s, speed: 130.94st/s, total perplexity: 6.9542
Epoch: 66 training finished. Time: 2883.78s, speed: 0.69st/s, total loss: 0
Epoch: 66 training finished. Time: 2883.78s, speed: 138.27st/s, total perplexity: 6.9826
gold_num = 5942 pred_num = 5920 right_num = 5564
gold_num = 2817 pred_num = 3205 right_num = 2084
Dev-Source: time: 15.08s, speed: 362.55st/s; acc: 0.9865, p: 0.9399, r: 0.9364, f: 0.9381
Test-Target: time: 15.08s, speed: 362.55st/s; acc: 0.9623, p: 0.6502, r: 0.7398, f: 0.6921
change optim sgd:
The best Source-domain dev f-score: 0.9390, Target-domain f-score: 0.7009
Epoch: 67/100
Learning rate is set as: 0.0034482758620689655
Epoch: 67 training finished. Time: 2897.08s, speed: 5.17st/s, total loss: 694.9905628766865
Epoch: 67 training finished. Time: 2897.08s, speed: 130.34st/s, total perplexity: 6.9865
Epoch: 67 training finished. Time: 2897.08s, speed: 0.69st/s, total loss: 0
Epoch: 67 training finished. Time: 2897.08s, speed: 137.63st/s, total perplexity: 6.9883
gold_num = 5942 pred_num = 5917 right_num = 5567
gold_num = 2817 pred_num = 3218 right_num = 2091
Dev-Source: time: 14.81s, speed: 369.24st/s; acc: 0.9867, p: 0.9408, r: 0.9369, f: 0.9389
Test-Target: time: 14.81s, speed: 369.24st/s; acc: 0.9626, p: 0.6498, r: 0.7423, f: 0.6930
change optim sgd:
The best Source-domain dev f-score: 0.9390, Target-domain f-score: 0.7009
Epoch: 68/100
Learning rate is set as: 0.0034090909090909085
Epoch: 68 training finished. Time: 2913.57s, speed: 5.14st/s, total loss: 703.0244072051719
Epoch: 68 training finished. Time: 2913.57s, speed: 129.60st/s, total perplexity: 6.9382
Epoch: 68 training finished. Time: 2913.57s, speed: 0.69st/s, total loss: 0
Epoch: 68 training finished. Time: 2913.57s, speed: 136.85st/s, total perplexity: 6.9262
gold_num = 5942 pred_num = 5919 right_num = 5570
gold_num = 2817 pred_num = 3226 right_num = 2096
Dev-Source: time: 15.90s, speed: 343.73st/s; acc: 0.9868, p: 0.9410, r: 0.9374, f: 0.9392
Test-Target: time: 15.90s, speed: 343.73st/s; acc: 0.9624, p: 0.6497, r: 0.7441, f: 0.6937
Exceed previous best f score: 0.9390336453326589
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9392, Target-domain f-score: 0.6937
Epoch: 69/100
Learning rate is set as: 0.0033707865168539322
Epoch: 69 training finished. Time: 3335.48s, speed: 4.49st/s, total loss: 694.337102599442
Epoch: 69 training finished. Time: 3335.48s, speed: 113.20st/s, total perplexity: 6.9760
Epoch: 69 training finished. Time: 3335.48s, speed: 0.60st/s, total loss: 0
Epoch: 69 training finished. Time: 3335.48s, speed: 119.54st/s, total perplexity: 6.9618
gold_num = 5942 pred_num = 5925 right_num = 5571
gold_num = 2817 pred_num = 3224 right_num = 2093
Dev-Source: time: 24.71s, speed: 221.50st/s; acc: 0.9867, p: 0.9403, r: 0.9376, f: 0.9389
Test-Target: time: 24.71s, speed: 221.50st/s; acc: 0.9622, p: 0.6492, r: 0.7430, f: 0.6929
change optim sgd:
The best Source-domain dev f-score: 0.9392, Target-domain f-score: 0.6937
Epoch: 70/100
Learning rate is set as: 0.003333333333333333
Epoch: 70 training finished. Time: 3224.21s, speed: 4.65st/s, total loss: 691.1552377054468
Epoch: 70 training finished. Time: 3224.21s, speed: 117.11st/s, total perplexity: 6.9500
Epoch: 70 training finished. Time: 3224.21s, speed: 0.62st/s, total loss: 0
Epoch: 70 training finished. Time: 3224.21s, speed: 123.67st/s, total perplexity: 6.9187
gold_num = 5942 pred_num = 5918 right_num = 5575
gold_num = 2817 pred_num = 3211 right_num = 2093
Dev-Source: time: 21.57s, speed: 253.58st/s; acc: 0.9869, p: 0.9420, r: 0.9382, f: 0.9401
Test-Target: time: 21.57s, speed: 253.58st/s; acc: 0.9625, p: 0.6518, r: 0.7430, f: 0.6944
Exceed previous best f score: 0.9392125453165838
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 71/100
Learning rate is set as: 0.0032967032967032963
Epoch: 71 training finished. Time: 3495.06s, speed: 4.29st/s, total loss: 677.995331185637
Epoch: 71 training finished. Time: 3495.06s, speed: 108.04st/s, total perplexity: 6.9162
Epoch: 71 training finished. Time: 3495.06s, speed: 0.57st/s, total loss: 0
Epoch: 71 training finished. Time: 3495.06s, speed: 114.09st/s, total perplexity: 6.8519
gold_num = 5942 pred_num = 5914 right_num = 5572
gold_num = 2817 pred_num = 3182 right_num = 2084
Dev-Source: time: 20.08s, speed: 272.24st/s; acc: 0.9869, p: 0.9422, r: 0.9377, f: 0.9399
Test-Target: time: 20.08s, speed: 272.24st/s; acc: 0.9623, p: 0.6549, r: 0.7398, f: 0.6948
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 72/100
Learning rate is set as: 0.0032608695652173916
Epoch: 72 training finished. Time: 3504.88s, speed: 4.28st/s, total loss: 689.0134121468291
Epoch: 72 training finished. Time: 3504.88s, speed: 107.73st/s, total perplexity: 6.9016
Epoch: 72 training finished. Time: 3504.88s, speed: 0.57st/s, total loss: 0
Epoch: 72 training finished. Time: 3504.88s, speed: 113.77st/s, total perplexity: 6.8838
gold_num = 5942 pred_num = 5924 right_num = 5577
gold_num = 2817 pred_num = 3227 right_num = 2094
Dev-Source: time: 19.77s, speed: 276.49st/s; acc: 0.9870, p: 0.9414, r: 0.9386, f: 0.9400
Test-Target: time: 19.77s, speed: 276.49st/s; acc: 0.9624, p: 0.6489, r: 0.7433, f: 0.6929
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 73/100
Learning rate is set as: 0.003225806451612903
Epoch: 73 training finished. Time: 3287.41s, speed: 4.56st/s, total loss: 686.5026794150472
Epoch: 73 training finished. Time: 3287.41s, speed: 114.86st/s, total perplexity: 6.9304
Epoch: 73 training finished. Time: 3287.41s, speed: 0.61st/s, total loss: 0
Epoch: 73 training finished. Time: 3287.41s, speed: 121.29st/s, total perplexity: 6.8630
gold_num = 5942 pred_num = 5931 right_num = 5572
gold_num = 2817 pred_num = 3240 right_num = 2094
Dev-Source: time: 16.46s, speed: 332.06st/s; acc: 0.9865, p: 0.9395, r: 0.9377, f: 0.9386
Test-Target: time: 16.46s, speed: 332.06st/s; acc: 0.9622, p: 0.6463, r: 0.7433, f: 0.6914
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 74/100
Learning rate is set as: 0.0031914893617021275
Epoch: 74 training finished. Time: 2910.32s, speed: 5.15st/s, total loss: 678.2296701157466
Epoch: 74 training finished. Time: 2910.32s, speed: 129.74st/s, total perplexity: 6.8831
Epoch: 74 training finished. Time: 2910.32s, speed: 0.69st/s, total loss: 0
Epoch: 74 training finished. Time: 2910.32s, speed: 137.01st/s, total perplexity: 6.8766
gold_num = 5942 pred_num = 5926 right_num = 5573
gold_num = 2817 pred_num = 3236 right_num = 2102
Dev-Source: time: 15.02s, speed: 363.79st/s; acc: 0.9868, p: 0.9404, r: 0.9379, f: 0.9392
Test-Target: time: 15.02s, speed: 363.79st/s; acc: 0.9623, p: 0.6496, r: 0.7462, f: 0.6945
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 75/100
Learning rate is set as: 0.003157894736842105
Epoch: 75 training finished. Time: 2920.30s, speed: 5.13st/s, total loss: 683.7522143269889
Epoch: 75 training finished. Time: 2920.30s, speed: 129.30st/s, total perplexity: 6.9034
Epoch: 75 training finished. Time: 2920.30s, speed: 0.68st/s, total loss: 0
Epoch: 75 training finished. Time: 2920.30s, speed: 136.54st/s, total perplexity: 6.8519
gold_num = 5942 pred_num = 5928 right_num = 5574
gold_num = 2817 pred_num = 3242 right_num = 2102
Dev-Source: time: 16.96s, speed: 322.32st/s; acc: 0.9868, p: 0.9403, r: 0.9381, f: 0.9392
Test-Target: time: 16.96s, speed: 322.32st/s; acc: 0.9622, p: 0.6484, r: 0.7462, f: 0.6938
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 76/100
Learning rate is set as: 0.0031249999999999993
Epoch: 76 training finished. Time: 2894.17s, speed: 5.18st/s, total loss: 679.2420950029045
Epoch: 76 training finished. Time: 2894.17s, speed: 130.47st/s, total perplexity: 6.8660
Epoch: 76 training finished. Time: 2894.17s, speed: 0.69st/s, total loss: 0
Epoch: 76 training finished. Time: 2894.17s, speed: 137.77st/s, total perplexity: 6.8711
gold_num = 5942 pred_num = 5923 right_num = 5569
gold_num = 2817 pred_num = 3221 right_num = 2077
Dev-Source: time: 14.41s, speed: 379.20st/s; acc: 0.9865, p: 0.9402, r: 0.9372, f: 0.9387
Test-Target: time: 14.41s, speed: 379.20st/s; acc: 0.9615, p: 0.6448, r: 0.7373, f: 0.6880
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 77/100
Learning rate is set as: 0.0030927835051546395
Epoch: 77 training finished. Time: 2908.32s, speed: 5.15st/s, total loss: 695.644865622744
Epoch: 77 training finished. Time: 2908.32s, speed: 129.83st/s, total perplexity: 6.8638
Epoch: 77 training finished. Time: 2908.32s, speed: 0.69st/s, total loss: 0
Epoch: 77 training finished. Time: 2908.32s, speed: 137.10st/s, total perplexity: 6.8309
gold_num = 5942 pred_num = 5928 right_num = 5574
gold_num = 2817 pred_num = 3235 right_num = 2090
Dev-Source: time: 18.70s, speed: 292.21st/s; acc: 0.9869, p: 0.9403, r: 0.9381, f: 0.9392
Test-Target: time: 18.70s, speed: 292.21st/s; acc: 0.9620, p: 0.6461, r: 0.7419, f: 0.6907
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 78/100
Learning rate is set as: 0.003061224489795918
Epoch: 78 training finished. Time: 2867.70s, speed: 5.23st/s, total loss: 669.6377582736313
Epoch: 78 training finished. Time: 2867.70s, speed: 131.67st/s, total perplexity: 6.8316
Epoch: 78 training finished. Time: 2867.70s, speed: 0.70st/s, total loss: 0
Epoch: 78 training finished. Time: 2867.70s, speed: 139.04st/s, total perplexity: 6.8656
gold_num = 5942 pred_num = 5928 right_num = 5578
gold_num = 2817 pred_num = 3219 right_num = 2075
Dev-Source: time: 14.45s, speed: 378.23st/s; acc: 0.9869, p: 0.9410, r: 0.9387, f: 0.9398
Test-Target: time: 14.45s, speed: 378.23st/s; acc: 0.9614, p: 0.6446, r: 0.7366, f: 0.6875
change optim sgd:
The best Source-domain dev f-score: 0.9401, Target-domain f-score: 0.6944
Epoch: 79/100
Learning rate is set as: 0.00303030303030303
Epoch: 79 training finished. Time: 2878.46s, speed: 5.21st/s, total loss: 677.3204550603405
Epoch: 79 training finished. Time: 2878.46s, speed: 131.18st/s, total perplexity: 6.8280
Epoch: 79 training finished. Time: 2878.46s, speed: 0.69st/s, total loss: 0
Epoch: 79 training finished. Time: 2878.46s, speed: 138.52st/s, total perplexity: 6.7707
gold_num = 5942 pred_num = 5912 right_num = 5573
gold_num = 2817 pred_num = 3196 right_num = 2084
Dev-Source: time: 17.38s, speed: 314.46st/s; acc: 0.9870, p: 0.9427, r: 0.9379, f: 0.9403
Test-Target: time: 17.38s, speed: 314.46st/s; acc: 0.9622, p: 0.6521, r: 0.7398, f: 0.6932
Exceed previous best f score: 0.9401349072512647
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 80/100
Learning rate is set as: 0.003
Epoch: 80 training finished. Time: 2870.59s, speed: 5.22st/s, total loss: 663.6177132418379
Epoch: 80 training finished. Time: 2870.59s, speed: 131.54st/s, total perplexity: 6.8070
Epoch: 80 training finished. Time: 2870.59s, speed: 0.70st/s, total loss: 0
Epoch: 80 training finished. Time: 2870.59s, speed: 138.90st/s, total perplexity: 6.7905
gold_num = 5942 pred_num = 5926 right_num = 5577
gold_num = 2817 pred_num = 3232 right_num = 2081
Dev-Source: time: 18.43s, speed: 296.63st/s; acc: 0.9869, p: 0.9411, r: 0.9386, f: 0.9398
Test-Target: time: 18.43s, speed: 296.63st/s; acc: 0.9618, p: 0.6439, r: 0.7387, f: 0.6880
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 81/100
Learning rate is set as: 0.0029702970297029703
Epoch: 81 training finished. Time: 2883.65s, speed: 5.20st/s, total loss: 661.1621418227442
Epoch: 81 training finished. Time: 2883.65s, speed: 130.94st/s, total perplexity: 6.8387
Epoch: 81 training finished. Time: 2883.65s, speed: 0.69st/s, total loss: 0
Epoch: 81 training finished. Time: 2883.65s, speed: 138.27st/s, total perplexity: 6.7998
gold_num = 5942 pred_num = 5920 right_num = 5570
gold_num = 2817 pred_num = 3208 right_num = 2085
Dev-Source: time: 18.45s, speed: 296.30st/s; acc: 0.9868, p: 0.9409, r: 0.9374, f: 0.9391
Test-Target: time: 18.45s, speed: 296.30st/s; acc: 0.9621, p: 0.6499, r: 0.7401, f: 0.6921
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 82/100
Learning rate is set as: 0.002941176470588235
Epoch: 82 training finished. Time: 2849.03s, speed: 5.26st/s, total loss: 667.8584471493959
Epoch: 82 training finished. Time: 2849.03s, speed: 132.53st/s, total perplexity: 6.8401
Epoch: 82 training finished. Time: 2849.03s, speed: 0.70st/s, total loss: 0
Epoch: 82 training finished. Time: 2849.03s, speed: 139.95st/s, total perplexity: 6.8151
gold_num = 5942 pred_num = 5931 right_num = 5573
gold_num = 2817 pred_num = 3231 right_num = 2093
Dev-Source: time: 14.28s, speed: 382.67st/s; acc: 0.9867, p: 0.9396, r: 0.9379, f: 0.9388
Test-Target: time: 14.28s, speed: 382.67st/s; acc: 0.9622, p: 0.6478, r: 0.7430, f: 0.6921
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 83/100
Learning rate is set as: 0.0029126213592233006
Epoch: 83 training finished. Time: 2712.65s, speed: 5.52st/s, total loss: 659.6833026856184
Epoch: 83 training finished. Time: 2712.65s, speed: 139.20st/s, total perplexity: 6.8023
Epoch: 83 training finished. Time: 2712.65s, speed: 0.74st/s, total loss: 0
Epoch: 83 training finished. Time: 2712.65s, speed: 146.99st/s, total perplexity: 6.8013
gold_num = 5942 pred_num = 5929 right_num = 5573
gold_num = 2817 pred_num = 3232 right_num = 2087
Dev-Source: time: 13.20s, speed: 414.15st/s; acc: 0.9868, p: 0.9400, r: 0.9379, f: 0.9389
Test-Target: time: 13.20s, speed: 414.15st/s; acc: 0.9617, p: 0.6457, r: 0.7409, f: 0.6900
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 84/100
Learning rate is set as: 0.0028846153846153843
Epoch: 84 training finished. Time: 2716.49s, speed: 5.52st/s, total loss: 649.846892288886
Epoch: 84 training finished. Time: 2716.49s, speed: 139.00st/s, total perplexity: 6.7775
Epoch: 84 training finished. Time: 2716.49s, speed: 0.74st/s, total loss: 0
Epoch: 84 training finished. Time: 2716.49s, speed: 146.78st/s, total perplexity: 6.8442
gold_num = 5942 pred_num = 5922 right_num = 5576
gold_num = 2817 pred_num = 3191 right_num = 2061
Dev-Source: time: 13.82s, speed: 395.44st/s; acc: 0.9870, p: 0.9416, r: 0.9384, f: 0.9400
Test-Target: time: 13.82s, speed: 395.44st/s; acc: 0.9614, p: 0.6459, r: 0.7316, f: 0.6861
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 85/100
Learning rate is set as: 0.002857142857142857
Epoch: 85 training finished. Time: 2723.12s, speed: 5.50st/s, total loss: 642.0847299844027
Epoch: 85 training finished. Time: 2723.12s, speed: 138.66st/s, total perplexity: 6.7902
Epoch: 85 training finished. Time: 2723.12s, speed: 0.73st/s, total loss: 0
Epoch: 85 training finished. Time: 2723.12s, speed: 146.43st/s, total perplexity: 6.8067
gold_num = 5942 pred_num = 5924 right_num = 5569
gold_num = 2817 pred_num = 3210 right_num = 2065
Dev-Source: time: 16.45s, speed: 332.33st/s; acc: 0.9866, p: 0.9401, r: 0.9372, f: 0.9386
Test-Target: time: 16.45s, speed: 332.33st/s; acc: 0.9615, p: 0.6433, r: 0.7330, f: 0.6852
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 86/100
Learning rate is set as: 0.002830188679245283
Epoch: 86 training finished. Time: 2701.04s, speed: 5.55st/s, total loss: 662.8191181011498
Epoch: 86 training finished. Time: 2701.04s, speed: 139.79st/s, total perplexity: 6.7902
Epoch: 86 training finished. Time: 2701.04s, speed: 0.74st/s, total loss: 0
Epoch: 86 training finished. Time: 2701.04s, speed: 147.62st/s, total perplexity: 6.7558
gold_num = 5942 pred_num = 5921 right_num = 5567
gold_num = 2817 pred_num = 3197 right_num = 2054
Dev-Source: time: 13.76s, speed: 397.20st/s; acc: 0.9866, p: 0.9402, r: 0.9369, f: 0.9385
Test-Target: time: 13.76s, speed: 397.20st/s; acc: 0.9613, p: 0.6425, r: 0.7291, f: 0.6831
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 87/100
Learning rate is set as: 0.002803738317757009
Epoch: 87 training finished. Time: 2796.02s, speed: 5.36st/s, total loss: 652.8355428059585
Epoch: 87 training finished. Time: 2796.02s, speed: 135.05st/s, total perplexity: 6.7575
Epoch: 87 training finished. Time: 2796.02s, speed: 0.71st/s, total loss: 0
Epoch: 87 training finished. Time: 2796.02s, speed: 142.61st/s, total perplexity: 6.7557
gold_num = 5942 pred_num = 5921 right_num = 5577
gold_num = 2817 pred_num = 3217 right_num = 2091
Dev-Source: time: 14.69s, speed: 372.11st/s; acc: 0.9870, p: 0.9419, r: 0.9386, f: 0.9402
Test-Target: time: 14.69s, speed: 372.11st/s; acc: 0.9623, p: 0.6500, r: 0.7423, f: 0.6931
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 88/100
Learning rate is set as: 0.0027777777777777775
Epoch: 88 training finished. Time: 3294.18s, speed: 4.55st/s, total loss: 660.9028896398377
Epoch: 88 training finished. Time: 3294.18s, speed: 114.62st/s, total perplexity: 6.7889
Epoch: 88 training finished. Time: 3294.18s, speed: 0.61st/s, total loss: 0
Epoch: 88 training finished. Time: 3294.18s, speed: 121.04st/s, total perplexity: 6.7151
gold_num = 5942 pred_num = 5913 right_num = 5571
gold_num = 2817 pred_num = 3194 right_num = 2086
Dev-Source: time: 17.49s, speed: 312.57st/s; acc: 0.9870, p: 0.9422, r: 0.9376, f: 0.9399
Test-Target: time: 17.49s, speed: 312.57st/s; acc: 0.9624, p: 0.6531, r: 0.7405, f: 0.6941
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 89/100
Learning rate is set as: 0.002752293577981651
Epoch: 89 training finished. Time: 3328.84s, speed: 4.50st/s, total loss: 641.1855686851777
Epoch: 89 training finished. Time: 3328.84s, speed: 113.43st/s, total perplexity: 6.7472
Epoch: 89 training finished. Time: 3328.84s, speed: 0.60st/s, total loss: 0
Epoch: 89 training finished. Time: 3328.84s, speed: 119.78st/s, total perplexity: 6.7561
gold_num = 5942 pred_num = 5924 right_num = 5574
gold_num = 2817 pred_num = 3225 right_num = 2090
Dev-Source: time: 18.13s, speed: 301.52st/s; acc: 0.9869, p: 0.9409, r: 0.9381, f: 0.9395
Test-Target: time: 18.13s, speed: 301.52st/s; acc: 0.9622, p: 0.6481, r: 0.7419, f: 0.6918
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 90/100
Learning rate is set as: 0.002727272727272727
Epoch: 90 training finished. Time: 3313.78s, speed: 4.52st/s, total loss: 630.2619496253319
Epoch: 90 training finished. Time: 3313.78s, speed: 113.95st/s, total perplexity: 6.7663
Epoch: 90 training finished. Time: 3313.78s, speed: 0.60st/s, total loss: 0
Epoch: 90 training finished. Time: 3313.78s, speed: 120.33st/s, total perplexity: 6.7303
gold_num = 5942 pred_num = 5917 right_num = 5572
gold_num = 2817 pred_num = 3200 right_num = 2076
Dev-Source: time: 19.54s, speed: 279.77st/s; acc: 0.9868, p: 0.9417, r: 0.9377, f: 0.9397
Test-Target: time: 19.54s, speed: 279.77st/s; acc: 0.9621, p: 0.6488, r: 0.7370, f: 0.6900
change optim sgd:
The best Source-domain dev f-score: 0.9403, Target-domain f-score: 0.6932
Epoch: 91/100
Learning rate is set as: 0.002702702702702703
Epoch: 91 training finished. Time: 3305.21s, speed: 4.53st/s, total loss: 635.3417489416897
Epoch: 91 training finished. Time: 3305.21s, speed: 114.24st/s, total perplexity: 6.7644
Epoch: 91 training finished. Time: 3305.21s, speed: 0.60st/s, total loss: 0
Epoch: 91 training finished. Time: 3305.21s, speed: 120.64st/s, total perplexity: 6.7078
gold_num = 5942 pred_num = 5923 right_num = 5579
gold_num = 2817 pred_num = 3221 right_num = 2079
Dev-Source: time: 17.32s, speed: 315.61st/s; acc: 0.9871, p: 0.9419, r: 0.9389, f: 0.9404
Test-Target: time: 17.32s, speed: 315.61st/s; acc: 0.9615, p: 0.6455, r: 0.7380, f: 0.6886
Exceed previous best f score: 0.9402733254597604
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9404, Target-domain f-score: 0.6886
Epoch: 92/100
Learning rate is set as: 0.002678571428571428
Epoch: 92 training finished. Time: 3131.71s, speed: 4.79st/s, total loss: 642.1352096521296
Epoch: 92 training finished. Time: 3131.71s, speed: 120.57st/s, total perplexity: 6.6824
Epoch: 92 training finished. Time: 3131.71s, speed: 0.64st/s, total loss: 0
Epoch: 92 training finished. Time: 3131.71s, speed: 127.32st/s, total perplexity: 6.7384
gold_num = 5942 pred_num = 5930 right_num = 5582
gold_num = 2817 pred_num = 3223 right_num = 2091
Dev-Source: time: 17.75s, speed: 307.85st/s; acc: 0.9869, p: 0.9413, r: 0.9394, f: 0.9404
Test-Target: time: 17.75s, speed: 307.85st/s; acc: 0.9616, p: 0.6488, r: 0.7423, f: 0.6924
change optim sgd:
The best Source-domain dev f-score: 0.9404, Target-domain f-score: 0.6886
Epoch: 93/100
Learning rate is set as: 0.0026548672566371677
Epoch: 93 training finished. Time: 2986.55s, speed: 5.02st/s, total loss: 644.086824413389
Epoch: 93 training finished. Time: 2986.55s, speed: 126.43st/s, total perplexity: 6.7226
Epoch: 93 training finished. Time: 2986.55s, speed: 0.67st/s, total loss: 0
Epoch: 93 training finished. Time: 2986.55s, speed: 133.51st/s, total perplexity: 6.8015
gold_num = 5942 pred_num = 5925 right_num = 5578
gold_num = 2817 pred_num = 3227 right_num = 2099
Dev-Source: time: 17.69s, speed: 308.94st/s; acc: 0.9868, p: 0.9414, r: 0.9387, f: 0.9401
Test-Target: time: 17.69s, speed: 308.94st/s; acc: 0.9617, p: 0.6504, r: 0.7451, f: 0.6946
change optim sgd:
The best Source-domain dev f-score: 0.9404, Target-domain f-score: 0.6886
Epoch: 94/100
Learning rate is set as: 0.002631578947368421
Epoch: 94 training finished. Time: 3136.23s, speed: 4.78st/s, total loss: 638.1244297917001
Epoch: 94 training finished. Time: 3136.23s, speed: 120.40st/s, total perplexity: 6.7299
Epoch: 94 training finished. Time: 3136.23s, speed: 0.64st/s, total loss: 0
Epoch: 94 training finished. Time: 3136.23s, speed: 127.14st/s, total perplexity: 6.7538
gold_num = 5942 pred_num = 5917 right_num = 5570
gold_num = 2817 pred_num = 3220 right_num = 2095
Dev-Source: time: 17.96s, speed: 304.31st/s; acc: 0.9867, p: 0.9414, r: 0.9374, f: 0.9394
Test-Target: time: 17.96s, speed: 304.31st/s; acc: 0.9619, p: 0.6506, r: 0.7437, f: 0.6941
change optim sgd:
The best Source-domain dev f-score: 0.9404, Target-domain f-score: 0.6886
Epoch: 95/100
Learning rate is set as: 0.002608695652173913
Epoch: 95 training finished. Time: 3152.39s, speed: 4.75st/s, total loss: 627.4881282187998
Epoch: 95 training finished. Time: 3152.39s, speed: 119.78st/s, total perplexity: 6.7552
Epoch: 95 training finished. Time: 3152.39s, speed: 0.63st/s, total loss: 0
Epoch: 95 training finished. Time: 3152.39s, speed: 126.49st/s, total perplexity: 6.7005
gold_num = 5942 pred_num = 5918 right_num = 5585
gold_num = 2817 pred_num = 3207 right_num = 2074
Dev-Source: time: 15.99s, speed: 341.79st/s; acc: 0.9872, p: 0.9437, r: 0.9399, f: 0.9418
Test-Target: time: 15.99s, speed: 341.79st/s; acc: 0.9617, p: 0.6467, r: 0.7362, f: 0.6886
Exceed previous best f score: 0.9404129793510325
Save current best model in file: data/conll03NER/lstm_crf_ner.model
change optim sgd:
The best Source-domain dev f-score: 0.9418, Target-domain f-score: 0.6886
Epoch: 96/100
Learning rate is set as: 0.0025862068965517237
Epoch: 96 training finished. Time: 3142.24s, speed: 4.77st/s, total loss: 632.3320020297542
Epoch: 96 training finished. Time: 3142.24s, speed: 120.17st/s, total perplexity: 6.7245
Epoch: 96 training finished. Time: 3142.24s, speed: 0.64st/s, total loss: 0
Epoch: 96 training finished. Time: 3142.24s, speed: 126.90st/s, total perplexity: 6.7088
gold_num = 5942 pred_num = 5918 right_num = 5577
gold_num = 2817 pred_num = 3206 right_num = 2085
Dev-Source: time: 17.05s, speed: 320.61st/s; acc: 0.9870, p: 0.9424, r: 0.9386, f: 0.9405
Test-Target: time: 17.05s, speed: 320.61st/s; acc: 0.9617, p: 0.6503, r: 0.7401, f: 0.6923
change optim sgd:
The best Source-domain dev f-score: 0.9418, Target-domain f-score: 0.6886
Epoch: 97/100
Learning rate is set as: 0.0025641025641025637
Epoch: 97 training finished. Time: 2778.10s, speed: 5.39st/s, total loss: 629.699469473213
Epoch: 97 training finished. Time: 2778.10s, speed: 135.92st/s, total perplexity: 6.7020
Epoch: 97 training finished. Time: 2778.10s, speed: 0.72st/s, total loss: 0
Epoch: 97 training finished. Time: 2778.10s, speed: 143.53st/s, total perplexity: 6.7322
gold_num = 5942 pred_num = 5924 right_num = 5582
gold_num = 2817 pred_num = 3221 right_num = 2102
Dev-Source: time: 13.16s, speed: 415.22st/s; acc: 0.9871, p: 0.9423, r: 0.9394, f: 0.9408
Test-Target: time: 13.16s, speed: 415.22st/s; acc: 0.9621, p: 0.6526, r: 0.7462, f: 0.6963
change optim sgd:
The best Source-domain dev f-score: 0.9418, Target-domain f-score: 0.6886