def train(self, logger, sess, batch_xs, epoch_idx, batch_idx, batch_total,
log_flag, keep_prob):
random_z = get_random_normal(batch_xs.shape[0], self.z_dim)
for i in range(5):
_, dis_loss_val = sess.run([self.dis_solver, self.dis_loss],
feed_dict={
self.X: batch_xs,
self.keep_prob: keep_prob,
self.z: random_z
})
_, dec_loss_val = sess.run([self.dec_solver, self.dec_loss],
feed_dict={
self.X: batch_xs,
self.keep_prob: keep_prob,
self.z: random_z
})
_, enc_loss_val = sess.run([self.enc_solver, self.enc_loss],
feed_dict={
self.X: batch_xs,
self.keep_prob: keep_prob
})
total_loss_val = dis_loss_val + dec_loss_val + enc_loss_val
if log_flag == True:
logger.debug(
'Epoch %.3i, Batch[%.3i/%i], Dis loss : %.4E, Dec loss : %.4E, Enc loss : %.4E, Train loss: %.4E'
% (epoch_idx, batch_idx + 1, batch_total, dis_loss_val,
dec_loss_val, enc_loss_val, total_loss_val))
return total_loss_val
def inference_with_output(self, logger, sess, batch_xs, epoch_idx, batch_idx, batch_total, log_flag, keep_prob):
random_z = get_random_normal(batch_xs.shape[0], self.z_dim)
dis_loss_val, dec_loss_val, output_val = sess.run([self.dis_loss, self.dec_loss, self.output], feed_dict={self.X: batch_xs, self.keep_prob: keep_prob, self.z: random_z})
total_loss_val = dis_loss_val + dec_loss_val
if log_flag == True:
logger.debug('Epoch %.3i, Batch[%.3i/%i], Dis loss : %.4E, Dec loss : %.4E, Test loss: %.4E' % (epoch_idx, batch_idx + 1, batch_total, dis_loss_val, dec_loss_val, total_loss_val))
return total_loss_val, output_val