def train(vocab_size, state_size, bptt_truncate, model_path, data_path,
num_epochs, learning_rate, model_dir):
# create an RNN, if possible load pre-existing model parameters
if model_path:
model_parameters = load_model_parameters(model_path)
model = RNN(vocab_size, state_size, bptt_truncate, model_parameters)
else:
model = RNN(vocab_size, state_size, bptt_truncate)
# construct datasets
training_data, validation_data, test_data = \
parse_reddit_data(vocab_size, data_path)
# train the model
model.sgd(training_data, num_epochs, learning_rate, validation_data,
test_data, model_dir)
