小编Vin*_*ier的帖子

我正在尝试编写一个脚本,允许我绘制一个数字图像,然后确定在MNIST上训练的模型的数字.

这是我的代码:

import random
import image
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
import numpy as np
import scipy.ndimage

mnist = input_data.read_data_sets( "MNIST_data/", one_hot=True )

x = tf.placeholder(tf.float32, [None, 784])
W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))

y = tf.nn.softmax(tf.matmul(x, W) + b)
y_ = tf.placeholder(tf.float32, [None, 10])

cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y), reduction_indices=[1]))

train_step = tf.train.GradientDescentOptimizer(0.5).minimize (cross_entropy)

init = tf.initialize_all_variables()

sess = tf.Session()
sess.run(init)

for i in range( 1000 ):
    batch_xs, batch_ys = mnist.train.next_batch( 1000 )
    sess.run(train_step, feed_dict= …