trying to understand bidirectional rnn

try_birnn.py

batch_size = 1
input_size = 2
num_units = 3

input_length = 8


sess = tf.Session()
with tf.variable_scope("root", initializer=tf.constant_initializer(0.5)) as scope:

  # our cells
  cell_fw = tf.nn.rnn_cell.LSTMCell(num_units=num_units, 
                           input_size=input_size)

  cell_bw = tf.nn.rnn_cell.LSTMCell(num_units=num_units, 
                           input_size=input_size)

  initializer = tf.random_uniform_initializer(-0.01, 0.01, seed=123)
  sequence_length = tf.placeholder(tf.int64)
  cell_fw = tf.nn.rnn_cell.LSTMCell(
      num_units, input_size, initializer=initializer)
  cell_bw = tf.nn.rnn_cell.LSTMCell(
      num_units, input_size, initializer=initializer)

  # input_seq = tf.placeholder(tf.float32, [input_length, None, input_size])
  input_seq = input_length * [tf.placeholder(tf.float32, shape=(batch_size, input_size))]


  outputs = tf.nn.bidirectional_rnn(
      cell_fw, cell_bw, input_seq, dtype=tf.float32,
      sequence_length=sequence_length)

  inputs_values = [np.random.randn(batch_size, input_size) for i in range(input_length)]

  sess.run([tf.initialize_all_variables()])

  feed_dict = {}
  for i in range(input_length):
    feed_dict[input_seq[i]] = inputs_values[i]
  feed_dict[sequence_length] = [5]

  res = sess.run([outputs], feed_dict=feed_dict)
  print(res)