小编Bat*_*tta的帖子

请考虑以下包含tensorflow的代码段tf.cond().

import tensorflow as tf
import numpy as np

bb = tf.placeholder(tf.bool)
xx = tf.placeholder(tf.float32, name='xx')
yy = tf.placeholder(tf.float32, name='yy')

zz = tf.cond(bb, lambda: xx + yy, lambda: 100 + yy)

with tf.Session() as sess:
        dict1 = {bb:False, yy:np.array([1., 3, 4]), xx:np.array([5., 6, 7])}
        print(sess.run(zz, feed_dict=dict1)) # works fine without errors

        dict2 = {bb:False, yy:np.array([1., 3, 4])}
        print(sess.run(zz, feed_dict=dict2)) # get an InvalidArgumentError asking to
                                             # provide an input for xx

在这两种情况下,bbis False和zz理论上的评估都没有依赖关系 …