Процесс прогнозирования Seq2Seq

Question

Я разработал модель кодера-декодера для прогнозирования временных рядов, но я не уверен, как достигается процесс прогнозирования.

Пример кода ниже.

  X = tf.placeholder(tf.float32,shape=(None,x_seq_len,n_features),name="X")
  y = tf.placeholder(tf.float32,shape=(None,y_seq_len,n_features),name="y")

  cells = tf.nn.rnn_cell.MultiRNNCell([ tf.nn.rnn_cell.LSTMCell(n_neurons) for _ in range(n_layers) ])

  init_state = cells.zero_state(batch_size, tf.float32)
  enc_outputs, enc_states = tf.nn.dynamic_rnn(cells, X, initial_state=init_state)

  dec_outputs,dec_states = tf.nn.dynamic_rnn(cells, y, initial_state=enc_states)

  loss = tf.reduce_mean(tf.square(dec_outputs - y))
  train_op = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss)

  #Training process
  ...      

  #Prediction process
  dec_outputs,dec_states = tf.nn.dynamic_rnn(cells, y, initial_state=enc_states)
  dec_states_prev = np.zeros((batch_size,seq_len,n_inputs))
  for x_test,y_test in test_set:
      dec_out,dec_states = sess.run([dec_o,dec_s],feed={y:x_test,dec_state:dec_states_prev})
      dec_states_prev = dec_states

      pred_arr.append(dec_out)

  #compare the predictions with targets(y_test).

Заранее спасибо.