Performance

This is for run performance report on models with bert-embedding.

Classification

from kashgari.corpus import SMP2018ECDTCorpus

train_x, train_y = SMP2018ECDTCorpus.load_data('train')
valid_x, valid_y = SMP2018ECDTCorpus.load_data('valid')
test_x, test_y = SMP2018ECDTCorpus.load_data('test')

	model_name	epoch	f1-score	precision	recall	time
0	BiGRU_Model	10	0.9335	0.937795	0.935065	00:33
1	BiLSTM_Model	10	0.929075	0.930548	0.92987	00:33
2	CNN_Attention_Model	10	0.862197	0.888507	0.866234	00:27
3	CNN_GRU_Model	10	0.840024	0.886519	0.850649	00:28
4	CNN_LSTM_Model	10	0.424649	0.551247	0.511688	00:27
5	CNN_Model	10	0.930336	0.938373	0.931169	00:26

NER

from kashgari.corpus import ChineseDailyNerCorpus

train_x, train_y = ChineseDailyNerCorpus.load_data('train')
valid_x, valid_y = ChineseDailyNerCorpus.load_data('valid')
test_x, test_y = ChineseDailyNerCorpus.load_data('test')

	model_name	epoch	f1-score	precision	recall	time
0	BiGRU_Model	10	0.921583	0.913184	0.930532	19:10
1	BiGRU_CRF_Model	10	0.935163	0.931246	0.939118	24:30
2	BiLSTM_Model	10	0.915363	0.906566	0.924418	19:12
3	BiLSTM_CRF_Model	10	0.940539	0.944549	0.936646	24:31
4	CNN_LSTM_Model	10	0.919783	0.909695	0.930272	19:07