RLANS

Learning Adversarial Networks for Semi-Supervised Text Classification via Policy Gradient

Require: Python 3.6, Pytorch 1.0

1)You should split the data into labeled and unlabeled data

python split_labeled_unlabeled.py

2)You should training the language model

python language_model_training.py --cuda --batch_size=32 --lr=0.01 --reduce_rate=0.9 --save='/ag_lm_model/'

the "--save" specifies the direction where you want save your pretrained language model.

3)You can conduct adv training for RLANS model

python Adversarial_training.py --cuda --lr=0.001 --batch_size=128 --save='/ag_adv_model/' --pre_train='/ag_lm_model' --number_per_class=1000 --reduce_rate=0.95

the "--pre_train" specifies the direction where your pretrained language model has saved.

the "--save" specifies the direction where you want save your RLANS model, and the testing result of each epoch.

This model refers to the paper:

@inproceedings{li2018learning,
  title={Learning adversarial networks for semi-supervised text classification via policy gradient},
  author={Li, Yan and Ye, Jieping},
  booktitle={Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining},
  pages={1715--1723},
  year={2018},
  organization={ACM}  
}

4)For comparison, you can also train a text classification model just with labeled, without specify the pretrained language model

python classifier_training.py --cuda --lr=0.001 --batch_size=128 --save='/classify_no_pre/' --pre_train='' --number_per_class=1000 --reduce_rate=0.95

the "--save" specifies the direction where you want save the LSTM based text classification model, and the testing result of each epoch.

4)For comparison, you can train the semisupervised sequence learning model (SSL), with the pretrained language model

python classifier_training.py --cuda --lr=0.001 --batch_size=128 --save='/classify_with_pre/' --pre_train='/ag_lm_model' --number_per_class=1000 --reduce_rate=0.95

the "--save" specifies the direction where you want save the SSL model, and the testing result of each epoch.

This model refers to the paper:

@inproceedings{dai2015semi,
  title={Semi-supervised sequence learning},
  author={Dai, Andrew M and Le, Quoc V},
  booktitle={Advances in neural information processing systems},
  pages={3079--3087},
  year={2015}
}

Note: this is a reproduced pytorch code to show the mode and ideals in the paper "Learning adversarial networks for semi-supervised text classification via policy gradient", which might not reproduce the results in the Table 3 of the paper. However, this can prove that the RLANS model outperforms the standard LSTM model and SSL model.