mnist手写数字识别—逻辑回归（batch|mnist手写数字识别—逻辑回归（batch 梯度下降法） mnist手写数字识别—逻辑回归（

1、导入所需包及训练、测试的数据

import tensorflow as tf import numpy as np import matplotlib.pyplot as plt from tensorflow.examples.tutorials.mnist import input_data mnist = input_data.read_data_sets('datasets', one_hot=True)

备注：自动下载所需数据到指定的文件夹，文件夹名称datasets
2、原始数据处理

trainimg = mnist.train.images trainlabel = mnist.train.labels testimg = mnist.test.images testlabel = mnist.test.labels

2-1、查看数据的形状

print(trainimg.shape) print(trainlabel.shape) print(testimg.shape) print(testlabel.shape)

output:
(55000, 784)
(55000, 10)
(10000, 784)
(10000, 10)
2-2、显示图片

i = 5 curr_img = np.reshape(trainimg[i,:], (28, 28)) plt.imshow(curr_img, cmap=plt.get_cmap('gray')) plt.show()

output:

文章图片

备注：
np.reshape() : 为数组提供新形状而不更改其数据
trainimg[i,:] : 提取数组第i行的全部数据 【mnist手写数字识别—逻辑回归（batch|mnist手写数字识别—逻辑回归（batch 梯度下降法）】
3、创建输入数据的特征向量

x = tf.placeholder(tf.float32,[None,784]) y = tf.placeholder(tf.float32,[None,10])

4、创建模型参数

W = tf.Variable(tf.zeros([784,10])) b = tf.Variable(tf.zeros([10]))

5、正向传播，计算输出值

actv = tf.nn.softmax(tf.matmul(x,W) + b)

6、计算损失函数

cost = tf.reduce_mean(-tf.reduce_sum(y*tf.log(tf.clip_by_value(actv,1e-10,1.0)) reduction_indices=1))

7、使用梯度下降法进行优化

learning_rate = 0.01 optm = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost)

8、判断正确结果与预测结果是否一致

pred = tf.equal(tf.argmax(actv, 1), tf.argmax(y, 1))

备注：
tf.argmax(input, dimension, name = None):返回最大数值的下标
dimension = 0: 按列找 ; dimension = 1 :按行找
9、计算正确率

accr = tf.reduce_mean(tf.cast(pred, tf.float32))

10、初始化模型参数

init_op = tf.global_variables_initializer()

11、定义常量

training_epochs = 100#训练的轮数 display_step = 5#用来比较、输出结果

12、使用batch梯度下降法进行训练

with tf.Session() as sess: sess.run(init_op) costs = [] #对于每一轮的训练 for epoch in range(training_epochs): avg_cost = 0.sess.run(optm, feed_dict={x: trainimg, y: trainlabel}) feeds = {x: trainimg, y: trainlabel} avg_cost += sess.run(cost, feed_dict=feeds) / training_epochsif epoch % display_step == 0: feed_train = {x: trainimg, y: trainlabel} feed_test = {x: testimg, y: testlabel}train_acc = sess.run(accr, feed_dict=feed_train) test_acc = sess.run(accr, feed_dict=feed_test)print("Eppoch: %03d/%03d cost: %.9f train_acc: %.3f test_acc: %.3f" % (epoch, training_epochs, avg_cost, train_acc, test_acc))costs.append(avg_cost)print("Done.")# plot the loss plt.plot(costs) plt.ylabel('cost') plt.xlabel('iterations ') plt.title("Learning rate =" + str(learning_rate)) plt.show()

output:

Eppoch: 000/100 cost: 0.022917860 train_acc: 0.663 test_acc: 0.670 Eppoch: 005/100 cost: 0.022393265 train_acc: 0.681 test_acc: 0.688 Eppoch: 010/100 cost: 0.021891801 train_acc: 0.695 test_acc: 0.700 Eppoch: 015/100 cost: 0.021410968 train_acc: 0.707 test_acc: 0.712 Eppoch: 020/100 cost: 0.020949018 train_acc: 0.717 test_acc: 0.722 Eppoch: 025/100 cost: 0.020504670 train_acc: 0.725 test_acc: 0.730 Eppoch: 030/100 cost: 0.020076952 train_acc: 0.730 test_acc: 0.737 Eppoch: 035/100 cost: 0.019665074 train_acc: 0.735 test_acc: 0.744 Eppoch: 040/100 cost: 0.019268372 train_acc: 0.740 test_acc: 0.749 Eppoch: 045/100 cost: 0.018886260 train_acc: 0.745 test_acc: 0.753 Eppoch: 050/100 cost: 0.018518188 train_acc: 0.749 test_acc: 0.757 Eppoch: 055/100 cost: 0.018163649 train_acc: 0.752 test_acc: 0.760 Eppoch: 060/100 cost: 0.017822148 train_acc: 0.755 test_acc: 0.763 Eppoch: 065/100 cost: 0.017493213 train_acc: 0.758 test_acc: 0.767 Eppoch: 070/100 cost: 0.017176378 train_acc: 0.761 test_acc: 0.773 Eppoch: 075/100 cost: 0.016871188 train_acc: 0.764 test_acc: 0.777 Eppoch: 080/100 cost: 0.016577202 train_acc: 0.767 test_acc: 0.780 Eppoch: 085/100 cost: 0.016293980 train_acc: 0.770 test_acc: 0.783 Eppoch: 090/100 cost: 0.016021105 train_acc: 0.772 test_acc: 0.786 Eppoch: 095/100 cost: 0.015758159 train_acc: 0.774 test_acc: 0.789 Done.