tensorflow搭建多层神经网络实现mnist识别

运行环境为Mac+python3.7，Mac安装tensorflow时一定要到GitHub上找到对应的版本。
数据集是mnist，在mnist官网进行下载，一共是四个压缩包，放入MNIST_data文件夹。训练集数据格式为[60000, 784]，测试集格式为[10000, 10]
初始版本为三层神经网络，神经元个数分别为784，50，10，激活函数为sigmoid，损失函数为softmax_cross_entropy_with_logits_v2，无正则化，优化器使用梯度下降法（SGD)，学习率为3.0，batch_size设置为50，达到的准确率接近97%

import tensorflow as tf
import numpy as np
from tensorflow.examples.tutorials.mnist import input_data
import matplotlib.pyplot as plt

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

inputSize = 784
outputSize = 10
hiddenSize = 50
batch_size = 50
epochs = 30
n_batch = mnist.train.num_examples // batch_size

x = tf.placeholder(tf.float32, shape=[None, inputSize])
y = tf.placeholder(tf.float32, shape=[None, outputSize])

# 创建一个简单的神经网络,输入层有784个特征，输出层有10个神经元：784-10
# 权值
W1 = tf.Variable(tf.truncated_normal([inputSize, hiddenSize], dtype=tf.float32, mean=0, stddev=0.1))
# 偏差
b1 = tf.Variable(tf.truncated_normal([hiddenSize]))
W2 = tf.Variable(tf.truncated_normal([hiddenSize, outputSize], dtype=tf.float32, mean=0, stddev=0.1))
b2 = tf.Variable(tf.truncated_normal([outputSize]))

# 使用sigmoid激活函数计算预测值
layer1 = tf.nn.sigmoid(tf.matmul(x, W1) + b1)
prediction = tf.nn.sigmoid(tf.matmul(layer1, W2) + b2)

# 损失函数
# loss = tf.losses.mean_squared_error(y, prediction)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits_v2(logits=prediction, labels=y)
loss = tf.reduce_mean(cross_entropy)

# 使用梯度下降法，学习率为3.0
train = tf.train.GradientDescentOptimizer(3.0).minimize(loss)


# 结果存放在一个布尔型列表中
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(prediction, 1))
# 求准确率
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

# 定义会话
with tf.Session() as sess:
    # 变量初始化
    sess.run(tf.global_variables_initializer())
    # 进行训练，周期epoch：所有数据训练一次，就是一个周期
    epos = []
    accs = []
    for epoch in range(epochs):
        for batch in range(n_batch):
            # 获取一个批次的数据和标签
            batch_xs, batch_ys = mnist.train.next_batch(batch_size)
            sess.run(train, feed_dict={x: batch_xs, y: batch_ys})
        # 每训练一个周期做一次测试，用测试集做测试
        acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels})
        epos.append(epoch)
        accs.append(acc)
        print("Epoch " + str(epoch) + ",Testing Accuracy " + str(acc))

    # 绘图
    plt.plot(epos, accs, 'b')
    plt.xlabel("epoch")
    plt.ylabel("accuracy")
    plt.title('Testing accuracy')
    plt.show()