Thank you for sharing the code from your paper!

You report the hyperparameter for your approach in the appendix and also in this repository. I was wondering whether this repository also contains the code which determined those hyper-parameters. Typically, you train many models and determine the best hyperparameter based on the best validation performance.

Slightly related to the hyper-parameters: in train.py you train the model with the reported hyper-parameter over many epochs and then evaluate the test performance. Based on the code it seems that you iterate over all model checkpoints (every n-th epoch), assess their performance on the test set, and then report the best test performance. Is that the code used to determine the test performance reported in the paper?

Would you please offer me "text_dict.pkl" ? I want to get the raw text,thank you!