tensorflow MNIST新手教程

官方教程代码如下：

 import gzip
 import os
 import tempfile  

 import numpy
 from six.moves import urllib
 from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 from tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets

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

 x = tf.placeholder(tf.float32,[None, 784]) #图像输入向量
 W = tf.Variable(tf.zeros([784,10]))  #权重，初始化值为全零
 b = tf.Variable(tf.zeros([10]))  #偏置，初始化值为全零  

 #进行模型计算，y是预测，y_ 是实际
 y = tf.nn.softmax(tf.matmul(x,W) + b)  

 y_ = tf.placeholder("float", [None,10])  

 #计算交叉熵
 cross_entropy = -tf.reduce_sum(y_*tf.log(y))
 #接下来使用BP算法来进行微调,以0.01的学习速率
 train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)  

 #上面设置好了模型，添加初始化创建变量的操作
 init = tf.global_variables_initializer()
 #启动创建的模型，并初始化变量
 sess = tf.Session()
 sess.run(init)
 #开始训练模型，循环训练1000次
 for i in range(1000):
     #随机抓取训练数据中的100个批处理数据点
     batch_xs, batch_ys = mnist.train.next_batch(100)
     sess.run(train_step, feed_dict={x:batch_xs,y_:batch_ys})  

 ''''' 进行模型评估 '''  

 #判断预测标签和实际标签是否匹配
 correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
 accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
 #计算所学习到的模型在测试数据集上面的正确率
 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

 import tensorflow as tf
 import struct
 import numpy as np

 with open('train-labels.idx1-ubyte','rb') as lb:
    magic,n=struct.unpack('>II',lb.read(8))
    labels = np.fromfile(lb,dtype=np.uint8)

 with open('train-images.idx3-ubyte','rb') as img:
    magic,num,rows,cols=struct.unpack('>IIII',img.read(16))
    images = np.fromfile(img,dtype=np.uint8).reshape(-1,784)
    images = images.astype(np.float32)
    images = np.multiply(images, 1.0 / 255.0)

 with open('t10k-labels.idx1-ubyte','rb') as lb:
    magic,n=struct.unpack('>II',lb.read(8))
    testlabels = np.fromfile(lb,dtype=np.uint8)

 with open('t10k-images.idx3-ubyte','rb') as img:
    magic,num,rows,cols=struct.unpack('>IIII',img.read(16))
    testimages = np.fromfile(img,dtype=np.uint8).reshape(-1,784)
    testimages = testimages.astype(np.float32)
    testimages = np.multiply(testimages, 1.0 / 255.0)

 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_sum(y_ * tf.log(y))
 train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)

 sess = tf.InteractiveSession()
 tf.global_variables_initializer().run()

 labels = sess.run(tf.one_hot(labels,10))
 testlabels = sess.run(tf.one_hot(testlabels,10))

 for _ in range(1000):
   a=np.random.permutation(np.arange(60000))
   bx = images[a[:100]]
   by = labels[a[:100]]
   sess.run(train_step,feed_dict={x:bx,y_:by})

 correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
 accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
 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