mnist 识别

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import os



def crete_weights(shape):
    # return tf.Variable(initial_value=tf.random_normal(shape=shape))
    return tf.Variable(initial_value=tf.random_normal(shape=shape, stddev = 0.01))# 初始化权重的 标准差

def crete_model(x):
    """
    构建卷积神经网络
    :return:
    """
    # 1 第一个卷积层
    with tf.variable_scope("conv1"):
        # 卷积层
        # 将x[None, 784]进行形状上的修改
        input_x = tf.reshape(x, shape = [-1, 28,28,1])
        # 定义32 个filter和偏置
        conv1_weights = crete_weights(shape=[5,5,1,32])
        conv1_bias = crete_weights(shape=[32])
        conv1_x = tf.nn.conv2d(input=input_x, filter=conv1_weights, strides=[1,1,1,1], padding="SAME") + conv1_bias
        # 激活层
        relu1_x = tf.nn.relu(conv1_x)

        # 池化层
        pool1_x = tf.nn.max_pool(value=relu1_x, ksize=[1,2,2,1], strides=[1,2,2,1], padding="SAME")
    # 1 第2个卷积层
    with tf.variable_scope("conv2"):
        # 卷积层
        # 定义32 个filter和偏置
        conv2_weights = crete_weights(shape=[5, 5, 32, 64])
        conv2_bias = crete_weights(shape=[64])
        conv2_x = tf.nn.conv2d(input=pool1_x, filter=conv2_weights, strides=[1, 1, 1, 1],
                               padding="SAME") + conv2_bias
        # 激活层
        relu2_x = tf.nn.relu(conv2_x)

        # 池化层
        pool2_x = tf.nn.max_pool(value=relu2_x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")
    # 全连接层
    with tf.variable_scope("full_connection"):
        x_fc = tf.reshape(pool2_x, shape=[-1,7*7*64])
        weights_fc = crete_weights(shape=[7*7*64,10])
        bias_fc = crete_weights(shape=[10])
        y_predict = tf.matmul(x_fc, weights_fc) + bias_fc
    # 调参  设置学习率  修改初始化权重的值 batch_normalization   dropout(防止过拟合)
    return y_predict

def full_connection():
    """
    用全连接对手写是数字进行识别
    :return:
    """
    # 1 准备数据
    mnist = input_data.read_data_sets('mnist_data', one_hot=True)
    with tf.variable_scope("mnist_data"):
        x  = tf.placeholder(dtype=tf.float32, shape=(None,784))
        y_true = tf.placeholder(dtype=tf.float32, shape=(None, 10))

    # with tf.variable_scope("fc_model"):
    #     weights  = tf.Variable(initial_value=tf.random_normal(shape=[784,10]))
    #     bias = tf.Variable(initial_value=tf.random_normal(shape=[10]))
    #     y_predict  = tf.matmul(x,weights) + bias
    y_predict = crete_model(x)
    # 第一个卷积大层：
    # 卷积层 32 filter
    # 激活层 relu
    # 赤化   大小 2 步长2
    # 第二同第一
    # 输入形状 [None, 28,28,1]
    # weights = tf.Variable(initial_value=tf.random_normal(shape=[5,5,1,32]))
    # bias = tf.Variable(initial_value=tf.random_normal(shape=[32]))
    # initial_value = random_normal(shape = [F,F,3,32])
    # 输出形状  [None, 28,28,32]
    # 3构造损失函数
    error = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_true, logits=y_predict))
    # 4 优化损失
    optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(error)# 一般是最多的
    # optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.1).minimize(error)

    # 5 准确率计算
    equal_list  = tf.equal(tf.argmax(y_true, 1),
                            tf.argmax(y_predict,1))# 返回的是bool型
    accuracy = tf.reduce_mean(tf.cast(equal_list, tf.float32))
    # 初始化变量
    init = tf.global_variables_initializer()
    # 开启会话
    with tf.Session() as sess:
        sess.run(init)
        image, label = mnist.train.next_batch(100)
        print("训练之前，损失为:%f" % sess.run(error, feed_dict={x:image, y_true:label}))
        # 开始训练
        for i in range(500):
            _, loss, accuracy_value  = sess.run([optimizer, error, accuracy], feed_dict={x:image, y_true:label})
            print("第%d次训练的损失为%f, 准确率为% f" %(i+1, loss, accuracy_value))
    return  None
# 准确率  比较输出的结果最大值与真实值最大值所在的位置
    # print(mnist.train.next_batch(1))
    # print(mnist.train.images[0])  # 查看特征值
    # print(mnist.train.labels[0])  # 目标值
    # one - hot  0 1 ...9

if __name__ == "__main__":
    full_connection()
    # crete_model()