Run as follows: python train.py -embeddings_type <> -dim <> -dataset <> -task <> -num_epochs <> -dropout <> -initial_lr <> values of arguments: -embeddings_type: lstm, glove or word2vec -dim: 300 for word2vec and glove, 1024 for lstm -dataset: verb_physics or PCE or special_words -task: three_way, four_way or one-pole -num_epochs: about 200 is enough -dropout: default is 0.5, recommend not to change -initial_lr: default is 0.05, may need to change to 0.1 or 0.01 additional flags -no_reverse: PCE(no-reverse) -remove_NA: Ignore examples with invalid comparisons -remove_sim: Ignore examples with similar comparisons -zero_shot: For Zero shot learning