link to IMDB dataset from Stanford: http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz
just copy the <train> directory from the downloaded file into the training_data folder inside the project.
The directory structure inside project will be:  training_data/train/{pos,neg,unsup}


The task is sentiment classification but the goal of the project is to show that
the CNN that comprises both branches of the siamese network is susceptible to adversial
examples and blind spots like those described in https://arxiv.org/abs/1312.6199 while the
siamese network built from the same CNN is not.