search

enricivi / adversarial_training_methods

Implementation of the methods proposed in **Adversarial Training Methods for Semi-Supervised Text Classification** on IMDB dataset (without pre-training)
MIT License
40 stars 11 forks source link

Parameters in the code #3

Closed lileicv closed 5 years ago

lileicv commented 5 years ago

The parameters you set are different from the original papers. Can you explain why?

For example, the p_mult is 5.0 in the paper. It is 0.02 in your code.

enricivi commented 5 years ago

Hi lileicv,

We had tried some parameters (but we did not do an hyperparameters fine-tuning) and this ones, for our implementation, seemed to work pretty well. It is also important to note that in this implementation we have removed the pre-training, using GloVe as word embedding. Thus it is possible that our hyperparameters are different compared to the paper's ones.

Enrico & Aurel