tensorflow-使用传统神经网络mnist手写数字图像识别的最佳实践

文件列表如下：

文件说明：mnist_inference 用于定义前向传播算法及其相关参数，mnist_train模型训练与持久化，mnist_eval模型加载与验证

mnist_inference.py

# 该文件定义了前向传播过程和神经网络参数
# _*_ coding:utf-8 _*_
import tensorflow as tf
#定义参数
INPUT_NODE = 784
OUTPUT_NODE = 10
LAYER1_NODE = 500
def get_weight_variable(shape,regularizer):
    #tf.get_variable用于变量的创建和加载，变量加载时可通过重命名直接使用滑动平均后的变量
    weights = tf.get_variable("weights",shape=shape,initializer=tf.truncated_normal_initializer(stddev=0.1))
    if regularizer != None:
        #losses为自定义集合名称
        tf.add_to_collection('losses',regularizer(weights))
    return  weights
#定义神经网络前向传播
def inference(input_tensor,regularizer):
    with tf.variable_scope('layer1'):
        weights = get_weight_variable([INPUT_NODE,LAYER1_NODE],regularizer)
        biases = tf.get_variable("biases",[LAYER1_NODE],initializer=tf.constant_initializer(0.0))
        layer1 = tf.nn.relu(tf.matmul(input_tensor,weights)+biases)

    with tf.variable_scope('layer2'):
        weights = get_weight_variable([LAYER1_NODE,OUTPUT_NODE], regularizer)
        biases = tf.get_variable("biases", [OUTPUT_NODE], initializer=tf.constant_initializer(0.0))
        layer2 = tf.nn.relu(tf.matmul(layer1, weights) + biases)
    return  layer2

 

mnist_train.py

# _*_ coding:utf-8 _*_
import os
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import mnist_inference

#配置神经网络参数
#使用梯度下降算法优化损失函数
BATCH_SIZE = 1000
#学习率指数衰减
LEARNING_RATE_BASE = 0.8
LEARNING_RATE_DECAY = 0.99
TRAINING_STEPS = 300000
#正则化避免过拟合
REGULARAZTION_RATE = 0.001
MOVING_AVERAGE_DECAY = 0.99

MODEL_SAVE_PATH = "save/"
MODEL_NAME="mnist.ckpt"

def train(mnist):
    x = tf.placeholder(tf.float32,[None,mnist_inference.INPUT_NODE],name='x-input')
    y_ = tf.placeholder(tf.float32, [None, mnist_inference.OUTPUT_NODE], name='y-input')
    regularizer = tf.contrib.layers.l2_regularizer(REGULARAZTION_RATE)
    y = mnist_inference.inference(x,regularizer)
    global_step = tf.Variable(0,trainable=False)
    variables_averages = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY,global_step)
    variables_averages_op = variables_averages.apply(tf.trainable_variables())
    #y为长度为10的1维数组，labels为正确答案的类别，所以用到了argmax
    cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y,labels=tf.argmax(y_,1))
    cross_entropy_mean = tf.reduce_mean(cross_entropy)
    loss = cross_entropy_mean + tf.add_n(tf.get_collection('losses'))
    print(mnist.train.num_examples)
    learning_rate = tf.train.exponential_decay(LEARNING_RATE_BASE,global_step,mnist.train.num_examples/BATCH_SIZE,LEARNING_RATE_DECAY)
    train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss,global_step=global_step)
    with tf.control_dependencies([train_step,variables_averages_op]):
        train_op = tf.no_op(name='train')
    saver = tf.train.Saver()
    with tf.compat.v1.Session() as sess:
        tf.global_variables_initializer().run()
        for i in range(TRAINING_STEPS):
            xs,ys = mnist.train.next_batch(BATCH_SIZE)
            _,loss_value,step = sess.run([train_op,loss,global_step],feed_dict={x:xs,y_:ys})
            if i % 1000 == 0:
                print("{0}steps,loss:{1}".format(step,loss_value))
                saver.save(sess,os.path.join(MODEL_SAVE_PATH,MODEL_NAME),global_step=global_step)
def main(argv=None):
    mnist = input_data.read_data_sets("data",one_hot=True)
    train(mnist)
if __name__ == '__main__':
    tf.app.run()

 

mnist_eval.py

# _*_ coding:utf-8_*_
import time
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import  mnist_inference,mnist_train
EVAL_INTERVAL_SECONDS = 10
def evaluate(mnist):
    with tf.Graph().as_default() as g:
        x = tf.placeholder(tf.float32, [None, mnist_inference.INPUT_NODE], name='x-input')
        y_ = tf.placeholder(tf.float32, [None, mnist_inference.OUTPUT_NODE], name='y-input')
        validate_feed = {x:mnist.validation.images,y_:mnist.validation.labels}
        #训练时不关心正则损失
        y = mnist_inference.inference(x,None)
        correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
        accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
        variables_averages = tf.train.ExponentialMovingAverage(mnist_train.MOVING_AVERAGE_DECAY)
        variables_to_restore = variables_averages.variables_to_restore()
        saver = tf.train.Saver(variables_to_restore)
        while True:
            with tf.compat.v1.Session() as sess:
                ckpt = tf.train.get_checkpoint_state(mnist_train.MODEL_SAVE_PATH)
                if ckpt and ckpt.model_checkpoint_path:
                    saver.restore(sess,ckpt.model_checkpoint_path)
                    global_step = ckpt.model_checkpoint_path.split('\\')[-1].split("-")[-1]
                    accuracy_score = sess.run(accuracy,feed_dict=validate_feed)
                    print("{0}steps,loss:{1}".format(global_step, accuracy_score))
            time.sleep(EVAL_INTERVAL_SECONDS)
def main(argv=None):
    mnist = input_data.read_data_sets("data",one_hot=True)
    evaluate(mnist)
if __name__ == '__main__':
    tf.app.run()