python - TensorFlow:如何通过复制张量之一来连接张量？

Question

我想通过复制其中一个张量来连接两个张量。例如，我有两个形状为 [2, 2, 3] 和 [2, 3] 的张量。结果应为 [2, 2, 6] 的形状。

t1 = [[[ 1, 1, 1], [2, 2, 2]],
      [[ 3, 3, 3], [4, 4, 4]]]
t2 = [[ 5, 5, 5], [6, 6, 6]]
"""
t3 = # some tf ops
t3 should be
t3 = [[[ 1, 1, 1, 5, 5, 5], [2, 2, 2, 5, 5, 5]],
      [[ 3, 3, 3, 6, 6, 6], [4, 4, 4, 6, 6, 6]]]
"""

因此，如果两个张量的形状为 [10, 5, 8] 和 [10, 3]，则结果的形状应为 [10, 5, 11]。

已更新

另一个例子:

t1 = np.reshape(np.arange(3*4*5), [3,4,5])
t2 = np.reshape(np.arange(3*1*2), [3,2])
""""
t3 should be
  [[[  0.,   1.,   2.,   3.,   4.,   0.,   1.],
    [  5.,   6.,   7.,   8.,   9.,   0.,   1.],
    [ 10.,  11.,  12.,  13.,  14.,   0.,   1.],
    [ 15.,  16.,  17.,  18.,  19.,   0.,   1.]],

   [[ 20.,  21.,  22.,  23.,  24.,   2.,   3.],
    [ 25.,  26.,  27.,  28.,  29.,   2.,   3.],
    [ 30.,  31.,  32.,  33.,  34.,   2.,   3.],
    [ 35.,  36.,  37.,  38.,  39.,   2.,   3.]],

   [[ 40.,  41.,  42.,  43.,  44.,   4.,   5.],
    [ 45.,  46.,  47.,  48.,  49.,   4.,   5.],
    [ 50.,  51.,  52.,  53.,  54.,   4.,   5.],
    [ 55.,  56.,  57.,  58.,  59.,   4.,   5.]]]
"""

Answer 1

函数tf.tile可以帮助你做到这一点。点击here获取函数的详细信息。

import numpy as np
import tensorflow as tf

t1 = np.reshape(np.arange(3*4*5), [3,4,5])
t2 = np.reshape(np.arange(3*1*2), [3,2])

# Keep t1 stay
t1_p = tf.placeholder(tf.float32, [3,4,5])

# Change t2 from shape(3,2) to shape(3,4,2) followed below two steps:
#    1. copy element of 2rd dimension of t2 as many times as you hope, as the updated example, it is 4
#    2. reshape the tiled tensor to compatible shape
t2_p = tf.placeholder(tf.float32, [3,2])
# copy the element of 2rd dimention of t2 by 4 times
t2_p_tiled = tf.tile(t2_p, [1, 4])
# reshape tiled t2 with shape(3,8) to the compatible shape(3,4,2)
t2_p_reshaped = tf.reshape(t2_p_tiled, [3,4,2])

# Concat t1 and changed t2, then you will get t3 you want
t3_p = tf.concat([t1_p, t2_p_reshaped], 2)

sess = tf.InteractiveSession()
t3 = sess.run(t3_p, {t1_p:t1, t2_p:t2})

print '*' * 20
print t1
print '*' * 20
print t2
print '*' * 20
print t3

# if you confused what the tf.tile did, you can print t2_p_tiled to see what happend
t2_tile = sess.run(t2_p_tiled, {t2_p:t2})
print '*' * 20
print t2_tile