Batch generator

[Yueeeeeeee]

Thanks for the great work! In the file batch_generator.py line 37, I noticed you have:

X = [self.inverse_map[y] + '=' + x for x, y in list(zip(X, Y))]

I'm wondering why you want to add self.inverse_map[y] + '=' to the text examples, and if that would leak label information in training? Thanks!

[jingyuanz]

hi, too long time ago i almost forgot but think the idea is to convert classification problem into matching problem so that you can get more training data. if you would concat k different labels with the sentence, you will have k times the training sample. the one concat with real label of course will trained to have max score. again, recommend to read protonet papers