官方教程代码如下：
1 import gzip 2 import os 3 import tempfile 4 5 import numpy 6 from six.moves import urllib 7 from six.moves import xrange # pylint: disable=redefined-builtin 8 import tensorflow as tf 9 from tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets 10 11 mnist = read_data_sets("MNIST_data/", one_hot=True) 12 13 x = tf.placeholder(tf.float32,[None, 784]) #图像输入向量 14 W = tf.Variable(tf.zeros([784,10])) #权重，初始化值为全零 15 b = tf.Variable(tf.zeros([10])) #偏置，初始化值为全零 16 17 #进行模型计算，y是预测，y_ 是实际 18 y = tf.nn.softmax(tf.matmul(x,W) + b) 19 20 y_ = tf.placeholder("float", [None,10]) 21 22 #计算交叉熵 23 cross_entropy = -tf.reduce_sum(y_*tf.log(y)) 24 #接下来使用BP算法来进行微调,以0.01的学习速率 25 train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy) 26 27 #上面设置好了模型，添加初始化创建变量的操作 28 init = tf.global_variables_initializer() 29 #启动创建的模型，并初始化变量 30 sess = tf.Session() 31 sess.run(init) 32 #开始训练模型，循环训练1000次 33 for i in range(1000): 34 #随机抓取训练数据中的100个批处理数据点 35 batch_xs, batch_ys = mnist.train.next_batch(100) 36 sess.run(train_step, feed_dict={x:batch_xs,y_:batch_ys}) 37 38 ''''' 进行模型评估 ''' 39 40 #判断预测标签和实际标签是否匹配 41 correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1)) 42 accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float")) 43 #计算所学习到的模型在测试数据集上面的正确率 44 print( sess.run(accuracy, feed_dict={x:mnist.test.images, y_:mnist.test.labels}) )
运行出现错误，不能导入数据，解决方案如下：
1..坑爹的GWF，严重阻碍人工智能的发展，与习大大十九大报告背道而驰，只能靠爱国青年曲线救国。方法为修改mnist.py文件（tensorflow.contrib.learn.python.learn.datasets.mnist），SOURCE_URL从 'https://storage.googleapis.com/cvdf-datasets/mnist/'改为 'http://yann.lecun.com/exdb/mnist/'，如此就能运行了。
2.手动下载 http://yann.lecun.com/exdb/mnist/，处理代码如下，注意一定要把图片像素除以255，否则正确率只有0.098
1 import tensorflow as tf 2 import struct 3 import numpy as np 4 5 with open('train-labels.idx1-ubyte','rb') as lb: 6 magic,n=struct.unpack('>II',lb.read(8)) 7 labels = np.fromfile(lb,dtype=np.uint8) 8 9 with open('train-images.idx3-ubyte','rb') as img: 10 magic,num,rows,cols=struct.unpack('>IIII',img.read(16)) 11 images = np.fromfile(img,dtype=np.uint8).reshape(-1,784) 12 images = images.astype(np.float32) 13 images = np.multiply(images, 1.0 / 255.0) 14 15 with open('t10k-labels.idx1-ubyte','rb') as lb: 16 magic,n=struct.unpack('>II',lb.read(8)) 17 testlabels = np.fromfile(lb,dtype=np.uint8) 18 19 with open('t10k-images.idx3-ubyte','rb') as img: 20 magic,num,rows,cols=struct.unpack('>IIII',img.read(16)) 21 testimages = np.fromfile(img,dtype=np.uint8).reshape(-1,784) 22 testimages = testimages.astype(np.float32) 23 testimages = np.multiply(testimages, 1.0 / 255.0) 24 25 x = tf.placeholder(tf.float32, [None, 784]) 26 W = tf.Variable(tf.zeros([784, 10])) 27 b = tf.Variable(tf.zeros([10])) 28 y = tf.nn.softmax(tf.matmul(x, W) + b) 29 y_ = tf.placeholder(tf.float32, [None, 10]) 30 cross_entropy = -tf.reduce_sum(y_ * tf.log(y)) 31 train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy) 32 33 sess = tf.InteractiveSession() 34 tf.global_variables_initializer().run() 35 36 labels = sess.run(tf.one_hot(labels,10)) 37 testlabels = sess.run(tf.one_hot(testlabels,10)) 38 39 for _ in range(1000): 40 a=np.random.permutation(np.arange(60000)) 41 bx = images[a[:100]] 42 by = labels[a[:100]] 43 sess.run(train_step,feed_dict={x:bx,y_:by}) 44 45 correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1)) 46 accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) 47 print(sess.run(accuracy, feed_dict={x:testimages, y_:testlabels}))
执行后结果约为0.92.
另外下列onehot方法非常好用：
def dense_to_one_hot(labels_dense, num_classes): """Convert class labels from scalars to one-hot vectors.""" num_labels = labels_dense.shape[0] index_offset = numpy.arange(num_labels) * num_classes labels_one_hot = numpy.zeros((num_labels, num_classes)) labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1 return labels_one_hot