Generates words character-by-character with an RNN given a target class (the inverse of torch-rnn-word-classifier)
th train.lua
-data_dir Data directory, containing a text file per class [data]
-hidden_size Hidden size of LSTM layer [200]
-dropout Dropout at last layer [0.5]
-learning_rate Learning rate [0.001]
-learning_rate_decay Learning rate decay [1e-07]
-max_length Maximum output length [20]
-n_epochs Number of epochs to train [100000]
$ th generate.lua -n_names 3
(French)
Armelle
Daneau
Tartan
(Japanese)
Ashi
Dano
Tokioki
(Italian)
Arlaninio
Dani
Tartani
(Chinese)
Ain
Dei
Ten