tensorflow逐元素乘法广播？

Question

tensorflow 是否为最后一维的元素乘法广播提供任何函数？

这是我正在尝试做的和不起作用的示例：

import tensorflow as tf
x = tf.constant(5, shape=(1, 200, 175, 6), dtype=tf.float32)
y = tf.constant(1, shape=(1, 200, 175), dtype=tf.float32)
tf.math.multiply(x, y)

基本上，我想对x 沿最后一个维度的切片中的每个切片，与y 进行逐元素矩阵乘法。

我发现这个问题要求类似的操作：Efficient element-wise multiplication of a matrix and a vector in TensorFlow

不幸的是，建议的方法（使用 tf.multiply() ）现在不再有效。相应的tf.math.multiply 也不起作用，因为上面的代码给了我下面的错误：

Traceback (most recent call last):
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 1864, in _create_c_op
    c_op = c_api.TF_FinishOperation(op_desc)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimensions must be equal, but are 175 and 200 for 'Mul' (op: 'Mul') with input shapes: [1,200,175,6], [1,200,175].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py", line 180, in wrapper
    return target(*args, **kwargs)
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/ops/math_ops.py", line 322, in multiply
    return gen_math_ops.mul(x, y, name)
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/ops/gen_math_ops.py", line 6490, in mul
    "Mul", x=x, y=y, name=name)
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/framework/op_def_library.py", line 788, in _apply_op_helper
    op_def=op_def)
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/util/deprecation.py", line 507, in new_func
    return func(*args, **kwargs)
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 3616, in create_op
    op_def=op_def)
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 2027, in __init__
    control_input_ops)
  File "/home/yuqiong/miniconda3/envs/deep/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 1867, in _create_c_op
    raise ValueError(str(e))
ValueError: Dimensions must be equal, but are 175 and 200 for 'Mul' (op: 'Mul') with input shapes: [1,200,175,6], [1,200,175].

我能想到一个可行的方法：复制y 6 次，使其具有与x 完全相同的形状，然后进行逐元素乘法。

但是在 tensorflow 中是否有更快且内存效率更高的方法？

Answer 1

这应该可以实现您想要的：

x = np.array([[[1,2,3],[4,5,6],[7,8,9],[10,11,12]]])
# [[[ 1  2  3]
#   [ 4  5  6]
#   [ 7  8  9]
#   [10 11 12]]]
y = np.array([[1,2,3,4]])
# [[1 2 3 4]]
y = tf.expand_dims(y, axis=-1)
mul = tf.multiply(x, y)
# [[[ 1  2  3]
#   [ 8 10 12]
#   [21 24 27]
#   [40 44 48]]]

最后，使用你需要的形状：

x = np.random.rand(1, 200, 175, 6)
y = np.random.rand(1, 200, 175)
y = tf.expand_dims(y, axis=-1)
mul = tf.multiply(x, y)
with tf.Session() as sess:
    print(sess.run(mul).shape)
    # (1, 200, 175, 6)
​