training options

model parameters

name	default	description
embedding_dim	100	the dimension of the embedding layer
hidden_dim	128	the dimension of the RNN hidden state
num_rnn_layer	1	the number of RNN layers
rnn_type	"lstm"	RNN type, choice: "lstm", "gru"
max_seq_len	100	max sequence length within training

training options

name	default	description
corpus_dir		the corpus directory
model_dir	"model_dir"	the output directory for model files
num_epoch	20	number of epoch to train
lr	1e-3	learning rate
weight_decay	0.0	the L2 normalization parameter
batch_size	1000	batch size
device	None	computing device: "cuda:0", "cpu:0". It will be autodetected by default
max_seq_len	100	max sequence length within training
val_split	0.2	the split for the validation dataset
test_split	0.2	the split for the testing dataset
recovery	False	continue to train from the saved model in model_dir
save_best_val_model	False	save the model whose validation score is smallest