adding ASGCN to the model list and refinement on the code

[GeneZC]

main changes:

• setting PYTHONHASHSEED to the fixed seed number.
• renaming a few variables to make them more distinguishable from each other.
• adding early stop on validation set with a patience number.
• adding support for all models available in an inference exemplar script, merging the inference script specifically designed for bert-based models.
• adding a graph-based model ASGCN.

test:

• ASGCN tested
• train tested
• train cross-val tested
• infer example tested

disclaimer:

• bert-based models are not tested

[yangheng95]

Great work!

[liana-minasyan]

Hello dears, I would like to train the models on a different TSA dataset without aspects but I get errors regarding the missing raw.graph files. I am wondering what those graphs are and whether or not it is possible to train on different datasets. Thank you very much!!!