tf.initialize_all_variables().run() mean_time_per_epoch = 0 start_time = time.time() print('\n' * 2) for epoch in range(conf.num_epochs): # additional variables to track the model losses = 0 batch_index = 0 # Model training current_learning_rate = conf.learning_rate * (conf.decay_rate**epoch) sess.run(tf.assign(train_model.learning_rate, current_learning_rate)) state = train_model.initial_state.eval() for batch_x, batch_y in reader.generateXYPairs(): # run_metadata = tf.RunMetadata() feed_dict = { train_model.input_x: batch_x, train_model.input_y: batch_y, train_model.initial_state: state, } summary, perplexity, cost_res, loss_res, state, _ = sess.run( [ train_model.merged, train_model.perplexity, train_model.cost, train_model.loss, train_model.final_state, train_model.train_operation ], feed_dict=feed_dict) losses += loss_res batch_index += 1