LSTM 变长

input为[2,10,8]的数组
batch = 2
steps = 10
为了变长 设置第二个batch的长度为6

import tensorflow as tf
import numpy as np
# 创建输入数据


cell = tf.contrib.rnn.BasicLSTMCell(num_units=4, state_is_tuple=True)
X = tf.placeholder(tf.float32,(2,10,8))
X_lengths = tf.placeholder(tf.float32)
outputs, last_states = tf.nn.dynamic_rnn(
    cell=cell,
    dtype=tf.float32,
    sequence_length=X_lengths,
    inputs=X)

with tf.Session() as sess:
    X1 = np.random.randn(2, 10, 8)
    # 第二个example长度为6
    X_lengths1 = [10, 6]
    sess.run(tf.global_variables_initializer())
    outputs, last_states = sess.run((outputs,last_states),feed_dict={X:X1,X_lengths:X_lengths1})
    print(last_states)
    print(outputs)

输出为
last state：[2,4]

LSTMStateTuple(c=array([[-0.43867487,  0.1597187 , -0.0061222 ,  0.21639484],
       [-0.34291208,  0.20014592,  0.5361769 ,  0.9636594 ]],
      dtype=float32), h=array([[-0.14935654,  0.08637159, -0.00271731,  0.14839597],
       [-0.17922135,  0.12582389,  0.25058392,  0.6113648 ]],
      dtype=float32))

outputs：[2,10,4]

[[[-0.02683123  0.02625443  0.16520582  0.13091983]
  [ 0.03294514 -0.0934423   0.19723308  0.16243443]
  [ 0.2861453   0.0485843   0.1357825   0.22259377]
  [-0.15690039  0.24221653 -0.09397997  0.02853295]
  [-0.35596934  0.21729332 -0.06792817 -0.10537033]
  [-0.1089786   0.02690388 -0.0379463  -0.09407488]
  [-0.05858612  0.02765152  0.10222299 -0.01180249]
  [-0.25069004  0.10124367  0.3203107  -0.00553809]
  [-0.12913188  0.15587549 -0.10340141 -0.03349639]
  [-0.14935654  0.08637159 -0.00271731  0.14839597]]

 [[ 0.02649872 -0.06729827 -0.2074864  -0.09794931]
  [-0.178599    0.07660254 -0.22099023 -0.19292895]
  [ 0.0912771  -0.10909905 -0.17696406 -0.05781336]
  [ 0.17243923 -0.17442465  0.03783391  0.17704639]
  [ 0.10158884 -0.150579    0.21208954  0.47918105]
  [-0.17922135  0.12582389  0.25058392  0.6113648 ]
  [ 0.          0.          0.          0.        ]
  [ 0.          0.          0.          0.        ]
  [ 0.          0.          0.          0.        ]
  [ 0.          0.          0.          0.        ]]]