两层卷积网络实现手写字母的识别（基于tensorflow）

可和这篇文章对比，https://blog.csdn.net/fanzonghao/article/details/81489049，数据集来源代码和链接一样。

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import read_pickle_dataset

train_dataset,train_label,valid_dataset,valid_label,test_dataset,test_label=read_pickle_dataset.pickle_dataset()
print('Training:', train_dataset.shape, train_label.shape)
print('Validing:', valid_dataset.shape, valid_label.shape)
print('Testing:', test_dataset.shape, test_label.shape)
print(valid_label[0])
"""
变化维度
"""
def reformat(dataset):
    dataset=dataset.reshape(-1,28,28,1).astype(np.float32)
    return dataset
"""
计算精度
"""
def accuracy(predictions,lables):
    acc=np.sum(np.argmax(predictions,1)==np.argmax(lables,1))/lables.shape[0]
    return acc
train_dataset=reformat(train_dataset)
valid_dataset=reformat(valid_dataset)
test_dataset=reformat(test_dataset)
print('Training:', train_dataset.shape, train_label.shape)
print('Validing:', valid_dataset.shape, valid_label.shape)
print('Testing:', test_dataset.shape, test_label.shape)

tf_valid_dataset=tf.constant(valid_dataset)
tf_test_dataset=tf.constant(test_dataset)
batch_size=128
def creat_placeholder():
    X = tf.placeholder(dtype=tf.float32, shape=(None, 28,28,1))
    Y = tf.placeholder(dtype=tf.float32, shape=(None, 10))
    return X,Y
"""
初始化权重：两层卷积
"""
def initialize_parameters():
    # CNN权重初始化
    W1 = tf.Variable(tf.truncated_normal(shape=[5,5,1,16],stddev=np.sqrt(2.0/(5*5*1))))
    b1=tf.Variable(tf.zeros(16))
    W2 = tf.Variable(tf.truncated_normal(shape=[5, 5, 16, 16], stddev=np.sqrt(2.0 / (5 * 5*16))))
    b2 = tf.Variable(tf.constant(1.0,shape=[16]))
    parameters={'W1':W1,
                'b1':b1,
                'W2':W2,
                'b2':b2
                }
    return parameters
def forward_propagation(X,parameters):
    W1 = parameters['W1']
    b1 = parameters['b1']
    W2 = parameters['W2']
    b2 = parameters['b2']

    Z1=tf.nn.conv2d(X,W1,strides=[1,1,1,1],padding='SAME')
    print('Z1.shape',Z1.shape)
    A1=tf.nn.relu(Z1+b1)
    P1 = tf.nn.max_pool(A1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
    print(P1.shape)
    Z2 = tf.nn.conv2d(P1, W2, strides=[1, 1, 1, 1], padding='SAME')
    print(Z2.shape)
    A2 = tf.nn.relu(Z2 + b2)
    P2 = tf.nn.max_pool(A2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
    print(P2.shape)

    P_flatten = tf.contrib.layers.flatten(P2)
    output = tf.contrib.layers.fully_connected(P_flatten, 10, activation_fn=None)

    return output
def compute_cost(Y,Y_pred):
#计算loss
    loss=tf.reduce_mean(                              #(10000,10)      (10000,10)
        tf.nn.softmax_cross_entropy_with_logits_v2(labels=Y,logits=Y_pred))
    return loss
def model():
    X,Y=creat_placeholder()
    parameters=initialize_parameters()
    Y_pred=forward_propagation(X, parameters)
    loss=compute_cost(Y, Y_pred)
    optimizer=tf.train.GradientDescentOptimizer(learning_rate=0.05).minimize(loss)
    # 预测
    train_prediction = tf.nn.softmax(Y_pred)
    init = tf.global_variables_initializer()

    corect_prediction = tf.equal(tf.argmax(Y_pred, 1), tf.argmax(Y, 1))
    accuarcy = tf.reduce_mean(tf.cast(corect_prediction, 'float'))
    with tf.Session() as sess:
        print('initialize')
        sess.run(init)
        costs = []
        for step in range(1001):
            offset = (step * batch_size) % (train_label.shape[0] - batch_size)
            batch_data = train_dataset[offset:(offset + batch_size), :, :, :]
            #batch_data = train_dataset[offset:(offset + batch_size), :]
            batch_label = train_label[offset:(offset + batch_size), :]
            feed = {X: batch_data, Y: batch_label}
            _, train_predictions, cost = sess.run([optimizer, train_prediction, loss], feed_dict=feed)
            if step%100==0:
                costs.append(cost)
                print('loss ={},at step {}'.format(cost, step))
                print('train accuracy={}'.format(accuracy(train_predictions, batch_label)))

                feed = {X: valid_dataset, Y: valid_label}
                valid_accuarcy = sess.run(accuarcy, feed_dict=feed)
                print('valid accuracy={}'.format(valid_accuarcy))
        feed = {X: test_dataset, Y: test_label}
        test_accuarcy = sess.run(accuarcy, feed_dict=feed)
        print('Test accuracy={}'.format(test_accuarcy))

    plt.plot(costs)
    plt.ylabel('cost')
    plt.xlabel('iterations ')
    plt.savefig('ELU.jpg')
    plt.show()
if __name__ == '__main__':
    model()