bert_cn

Chinese sentiment classification based on bert. This code was tested with TensorFlow 1.11.0 and Python 3.6.5

experiment results (comparing different optimizer)

Hyperparameters:

max_seq_length=128, train_batch_size=32, learning_rate=1e-6, num_train_epochs=50

Training Summary:

index	platform	device	optimizer	time	loss	accuracy
1	tensorflow	gpu	special adam	1h	0.9473789	0.815
2	tensorflow	gpu	adam	0.5h	0.4259515	0.825

Name	Name	Last commit message	Last commit date
Latest commit LinkToPast1990 Update README.md Nov 19, 2019 192f773 · Nov 19, 2019 History 14 Commits
data	data	add corpus and data_processing.py	Mar 13, 2019
model	model	Update README.md	Mar 13, 2019
output	output	add model and output	Mar 13, 2019
LICENSE	LICENSE	Create LICENSE	Mar 13, 2019
README.md	README.md	Update README.md	Nov 19, 2019
data_processing.py	data_processing.py	update data_processing.py	Mar 13, 2019
modeling.py	modeling.py	Initial	Mar 13, 2019
modeling_adam.py	modeling_adam.py	Initial	Mar 13, 2019
optimization.py	optimization.py	Initial	Mar 13, 2019
optimization_adam.py	optimization_adam.py	Initial	Mar 13, 2019
run_classifier_cn.py	run_classifier_cn.py	Initial	Mar 13, 2019
run_classifier_cn_adam.py	run_classifier_cn_adam.py	Initial	Mar 13, 2019
tokenization.py	tokenization.py	Initial	Mar 13, 2019