Implementation of the paper Gated-Attention Readers for Text Comprehension in PyTorch.
This code only supports training and validation on the CNN and Daily Mail cloze style dataset.
Data can be dowloaded here
Include all training data, validation data and testing data in the same folder, first training data followed by validation and testing data.
For training and testing the model run
python train.py
Optional Arguments
--epochs number of epochs
--iterations iterations per epoch
--lr learning rate for Adam optimizer
--char_size size for character embeddings
--embed_size size for glove embeddings
--char_hidden_size hidden size for character GRU
--hidden_size hidden size for document and query GRU
--use_char whether to use character embeddings or not
--use_features whether to use qe-comm features or not
--batch_size batch size for training
--gru_layers number of GRU layers for document and query
--train_file file having all the data (training+validation+testing)
--training_size number of training examples
--dev_size number og validation examples
--test_size number of testing examples