For CS 5984 Security Analytics project work

It is an experiment to check whether adversarial training is transferable in the text domain. So, the basic idea is if we train a model adversarially on adversarial samples from attack A, does it make the model more robust against attack B?

To run this project, you need to do the following things:

1. Create a file config.yml in the project root directory and copy the config.yml.demo file contents to the new file

2. Change the file paths if need be

3. Create folders named adv_samples, models, details, result inside the AdversaialTraining directory.

4. Open the AdversarialTraining/main.py (entry point) file and change the TODOs according to the following

   a. For training, set the task variable to "train"

   i. if you want to train your model normally, set **at** to False, specify the normal training parameters in Args constructor
   
   ii. if you want to train your model adversarially, set **at** to True, specify the adversarial training parameters in Args constructor
   

   b. For generating adversarial samples, set the task variable to "pre-generate", set at to False, specify the parameters for pre-generate in Args constructor, specify the model path that you want to attack

   c. For evaluating model's performance, set the task variable to "evaluate", set at to True or False depending on whether you want to evaluate a normal model or an adversarial model.

Reference: https://github.com/QData/TextAttack