TensorFlow在MNIST中的应用-卷积神经网络CNN

参考：

《TensorFlow技术解析与实战》

########################################################################################

用TensorFlow搭建一个卷积神经网络CNN模型，并用来训练MNIST数据集。

# -*- coding:utf-8 -*-# ==============================================================================# 20171115# HelloZEX# 卷积神经网络# ==============================================================================import tensorflow as tfimport numpy as npfrom tensorflow.examples.tutorials.mnist import input_data#定义训练和评估时的批次大小batch_size = 128test_size = 256#初始化权重函数def init_weights(shape):    return tf.Variable(tf.random_normal(shape, stddev=0.01))#神经网络模型的构建，传入以下参数# x：输入数据# w：每一层的权重# p_keep_conv，p_keep_hidden；dropout要保留的神经元比例def model(X, w, w2, w3, w4, w_o, p_keep_conv, p_keep_hidden):    #第一组卷积和池化层，最后dropout一些神经元    l1a = tf.nn.relu(tf.nn.conv2d(X, w,                       # l1a shape=(?, 28, 28, 32)                        strides=[1, 1, 1, 1], padding='SAME'))    l1 = tf.nn.max_pool(l1a, ksize=[1, 2, 2, 1],              # l1 shape=(?, 14, 14, 32)                        strides=[1, 2, 2, 1], padding='SAME')    l1 = tf.nn.dropout(l1, p_keep_conv)    #第二组卷积和池化层，最后dropout一些神经元    l2a = tf.nn.relu(tf.nn.conv2d(l1, w2,                     # l2a shape=(?, 14, 14, 64)                        strides=[1, 1, 1, 1], padding='SAME'))    l2 = tf.nn.max_pool(l2a, ksize=[1, 2, 2, 1],              # l2 shape=(?, 7, 7, 64)                        strides=[1, 2, 2, 1], padding='SAME')    l2 = tf.nn.dropout(l2, p_keep_conv)    #第三组卷积和池化层，最后dropout一些神经元    l3a = tf.nn.relu(tf.nn.conv2d(l2, w3,                     # l3a shape=(?, 7, 7, 128)                        strides=[1, 1, 1, 1], padding='SAME'))    l3 = tf.nn.max_pool(l3a, ksize=[1, 2, 2, 1],              # l3 shape=(?, 4, 4, 128)                        strides=[1, 2, 2, 1], padding='SAME')    l3 = tf.reshape(l3, [-1, w4.get_shape().as_list()[0]])    # reshape to (?, 2048)    l3 = tf.nn.dropout(l3, p_keep_conv)    #全连接层，最后dropout一些神经元    l4 = tf.nn.relu(tf.matmul(l3, w4))    l4 = tf.nn.dropout(l4, p_keep_hidden)    #输出层    pyx = tf.matmul(l4, w_o)    return pyx#得到训练和测试的图片mnist = input_data.read_data_sets("MNIST_Labels_Images", one_hot=True)trX, trY, teX, teY = mnist.train.images, mnist.train.labels, mnist.test.images, mnist.test.labelstrX = trX.reshape(-1, 28, 28, 1)  # 28x28x1 input imgteX = teX.reshape(-1, 28, 28, 1)  # 28x28x1 input imgp_keep_hidden: 1.0})))X = tf.placeholder("float", [None, 28, 28, 1])Y = tf.placeholder("float", [None, 10])#初始化权重w = init_weights([3, 3, 1, 32])       # patch大小为3x3，输入维度为1 ,输出维度为32w2 = init_weights([3, 3, 32, 64])     # patch大小为3x3，输入维度为32 ,输出维度为64w3 = init_weights([3, 3, 64, 128])    # patch大小为3x3，输入维度为64 ,输出维度为128w4 = init_weights([128 * 4 * 4, 625]) # 全连接层w_o = init_weights([625, 10])         # 输出层，输入维度为625，输出维度为10代表十个分类（labels）#我们定义dropout的占位符 keep_conv，他表示在一层中有多少比例的神经元被保留下来。生成网络模型，得到预测值p_keep_conv = tf.placeholder("float")p_keep_hidden = tf.placeholder("float")py_x = model(X, w, w2, w3, w4, w_o, p_keep_conv, p_keep_hidden)#定义的损失函数，并作均值处理。采用实现RMSProp算法的优化器cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=py_x, labels=Y))train_op = tf.train.RMSPropOptimizer(0.001, 0.9).minimize(cost)#定义预测的操作（predict_op）predict_op = tf.argmax(py_x, 1)# Launch the graph in a sessionwith tf.Session() as sess:    # you need to initialize all variables    tf.global_variables_initializer().run()    for i in range(100):        training_batch = zip(range(0, len(trX), batch_size), range(batch_size, len(trX)+1, batch_size))        for start, end in training_batch:            sess.run(train_op, feed_dict={X: trX[start:end], Y: trY[start:end], p_keep_conv: 0.8, p_keep_hidden: 0.5})        test_indices = np.arange(len(teX)) # Get A Test Batch        np.random.shuffle(test_indices)        test_indices = test_indices[0:test_size]        print(i, np.mean(np.argmax(teY[test_indices], axis=1) ==                         sess.run(predict_op, feed_dict={X: teX[test_indices],                                                         Y: teY[test_indices],                                                         p_keep_conv: 1.0,                                                         p_keep_hidden: 1.0})))

########################################################################################

/usr/bin/python2.7 /home/zhengxinxin/Desktop/PyCharm/Spark/SparkMNIST/SparkMNIST_CNN.pyExtracting MNIST_Labels_Images/train-images-idx3-ubyte.gzExtracting MNIST_Labels_Images/train-labels-idx1-ubyte.gzExtracting MNIST_Labels_Images/t10k-images-idx3-ubyte.gzExtracting MNIST_Labels_Images/t10k-labels-idx1-ubyte.gz2017-11-15 09:55:12.123463: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.2017-11-15 09:55:12.123492: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.2017-11-15 09:55:12.123497: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.2017-11-15 09:55:12.123500: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.2017-11-15 09:55:12.123503: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.(0, 0.9453125)(1, 0.97265625)(2, 0.98828125)(3, 0.984375)(4, 0.98046875)(5, 1.0)(6, 0.984375)(7, 0.99609375)(8, 0.9921875)(9, 0.99609375)(10, 0.99609375)(11, 0.99609375)(12, 0.98828125)(13, 0.99609375)(14, 1.0)(15, 0.98046875)(16, 0.9921875)(17, 0.99609375)(18, 0.99609375)(19, 1.0)(20, 0.98828125)(21, 0.9921875)(22, 0.98828125)(23, 1.0)(24, 1.0)(25, 0.98828125)(26, 1.0)(27, 0.9921875)(28, 0.9921875)(29, 0.9921875)(30, 0.99609375)(31, 0.99609375)(32, 1.0)(33, 1.0)(34, 0.9921875)(35, 0.99609375)(36, 1.0)(37, 0.9921875)(38, 0.984375)(39, 0.99609375)(40, 0.9921875)(41, 0.98828125)(42, 0.98828125)(43, 1.0)(44, 1.0)(45, 0.9921875)(46, 1.0)(47, 1.0)(48, 0.98828125)(49, 0.9921875)(50, 0.99609375)(51, 0.9921875)(52, 0.9921875)(53, 0.98828125)(54, 0.98828125)(55, 0.98828125)(56, 0.98828125)(57, 0.9921875)(58, 0.99609375)(59, 0.99609375)(60, 0.984375)(61, 0.99609375)(62, 0.99609375)(63, 0.99609375)(64, 0.99609375)(65, 0.9921875)(66, 0.99609375)(67, 0.99609375)(68, 0.9765625)(69, 0.99609375)(70, 0.9921875)(71, 0.9921875)(72, 0.99609375)(73, 0.9921875)(74, 0.9921875)(75, 0.9921875)(76, 0.98828125)(77, 0.99609375)(78, 0.99609375)(79, 0.99609375)(80, 0.984375)(81, 0.9921875)(82, 0.9921875)(83, 0.98828125)(84, 0.9765625)(85, 0.99609375)(86, 1.0)(87, 1.0)(88, 0.984375)(89, 0.99609375)(90, 0.9921875)(91, 0.9921875)(92, 0.984375)(93, 0.9921875)(94, 0.99609375)(95, 1.0)(96, 0.99609375)(97, 1.0)(98, 0.99609375)(99, 0.984375)Process finished with exit code 0

########################################################################################

上面输出了训练的次数和准确度的关系。可以看到100轮后准确度已经非常高了。通过回归模型和卷积神经网络模型，可以看出卷积神经网络的效果非常好。下一节使用RNN训练MNIST。