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