未调用张量流映射函数 [英] tensorflow map function not being invoked

问题描述

根据我对 tensorflow 中 map 函数的理解，我预计 my_map 会被调用 60,000 次，但它只被调用了一次.

based on my understanding of map function in tensorflow, I expected that my_map would be invoked 60,000 times but it was invoked only once.

输出

使用 TensorFlow 后端.(60000, 28, 28)Tensor("map/while/TensorArrayReadV3:0", shape=(28, 28), dtype=uint8)

Using TensorFlow backend. (60000, 28, 28) Tensor("map/while/TensorArrayReadV3:0", shape=(28, 28), dtype=uint8)

进程结束，退出代码 0

Process finished with exit code 0

代码:

import tensorflow as tf
from keras.datasets import mnist

(train_images, train_labels), (test_images, test_labels) = mnist.load_data()


def my_map(elem):
    print(elem)
    return elem


print(train_images.shape)

tf_map = tf.map_fn(fn=my_map, elems=train_images)


with tf.Session() as sess:
    sess.run(tf_map)

我做错了什么?任何帮助，将不胜感激.

What did I do wrong? Any help would be appreciated.

推荐答案

my_map 中的打印将无法用于打印.请测试一下:

The print in the my_map won't work for printing. Please, test this:

import tensorflow as tf
from keras.datasets import mnist

(train_images, train_labels), (test_images, test_labels) = mnist.load_data()


def my_map(elem):
    #print(elem)
    elem = elem + 1
    return elem


print(train_images.shape)

tf_map = tf.map_fn(fn=my_map, elems=train_images)


with tf.Session() as sess:
    print(sess.run(tf_map[0,0]))

这篇关于未调用张量流映射函数的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！