python - 如何在 Tensorflow 中为数组创建 np.subtract.outer

Question

假设x是一个二维数组，其中X1 = [a,b]并且a,b只是标量

然后输入如下:x = [X1, X2, X3]

我想计算:

y = [ [X1-X1, X1-X2, X1-X3],
      [X2-X1, X2-X2, X2-X3],
      [X3-X1, X3-X2, X3-X3] ]

Answer 1

这是使用隐式广播来实现此目的的一种方法:

import tensorflow as tf
import numpy as np

X_np = np.array([[1, 2], [3, 4], [10, 10]])

X_tf = tf.placeholder(tf.float32, shape=(None, None,))
y_tf = X_tf[:, None, :] - X_tf[None, :, :]

with tf.Session() as sess:
    out = sess.run(y_tf, feed_dict={X_tf: X_np})
    print(out)  # prints [[[ 0.  0.]
                #          [-2. -2.]
                #          [-9. -8.]]
                #          
                #         [[ 2.  2.]
                #          [ 0.  0.]
                #          [-7. -6.]]
                #          
                #         [[ 9.  8.]
                #          [ 7.  6.]
                #          [ 0.  0.]]]