1、关于数据集提取中遇到的问题：
1）情况如下||
实现数据集提取的代码函数部分如下
def reconstruct_image():
for iin range(10):
if not os.path.exists('./{}'.format(i)):创建一个文件目录
os. makedirs('./{}'.format(i))
batch_size=1
for i in range(int( mnist.train.num_examples / batch_size)):
# x_data, y_data = mnist.train.next_batch(batch_size)
x_data, y_data = mnist.train.next_batch(batch_size)
img = Image.fromarray(np.reshape(np.array(x_data[0]*255, dtype="uint8"), newshape=(28,28)))
dir = np.argmax(y_data[0])
img.save('./{}/{}.bmp'.format(dir, i)保存到相应的位置中去
在过程中，应该要十分的注意这个环节|：
1）在实现手写体识别的时候，添加层的函数，在tensorboard中，有可视化的工具，进行卷积，计算，池化，再进行卷积，再过来就是搭建网络，随之就是训练网络的具体的代码实现如下：
#/usr/bin/python3
#-*- coding:utf-8 -*-
#Author: jackxing
#@Time: 2018/11/7
import tensorflowas tf
from tensorflow.examples.tutorials.mnistimport input_data
import os
from PILimport Image
import numpyas np
mnist=input_data.read_data_sets('./MNIST', one_hot=True)
x=tf.placeholder(dtype=tf.float32, shape=[None,784], name='x')
y=tf.placeholder(dtype=tf.float32, shape=[None,10], name='y')
batch_size=1000
#添加层的函数
def add_layer(input_data, input_num, out_num, activation_function=None):
w=tf.Variable(initial_value=tf.random_normal(shape=[input_num, out_num]))
b=tf.Variable(initial_value=tf.random_normal(shape=[1, out_num]))
output=tf.add(tf.matmul(input_data, w), b)
if activation_function:
output=activation_function(output)
return output
#搭建网络的结构
def build_nn(data):
hidden_layer1=add_layer(data, 784, 100, activation_function=tf.nn.sigmoid)
hidden_layer2=add_layer(hidden_layer1, 100, 50, activation_function=tf.nn.sigmoid)
out_layer=add_layer(hidden_layer2, 50, 10)
return out_layer
#训练网络的结构
def train_nn(data):
output = build_nn(data)
loss=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=output))
optimizer=tf.train.GradientDescentOptimizer(learning_rate=1).minimize(loss)
saver=tf.train.Saver()
with tf.Session()as sess:
sess.run(tf.global_variables_initializer())
if not os.path.exists('checkpoint'):
for iin range(50):
epoch_cost =0
for _in range(int(mnist.train.num_examples /batch_size)):
x_data, y_data = mnist.train.next_batch(batch_size)
cost, _ = sess.run([loss, optimizer], feed_dict={x: x_data, y: y_data})
epoch_cost += cost
print('Ecoch', i, ':', epoch_cost)
accuracy = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(y, 1),tf.argmax(output, 1)), tf.float32))
acc = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels})
print(acc)
saver.save(sess, './mnist.ckpt')
else:
saver.restore(sess, './mnist.ckpt')
predict('./2.jpg', sess, output)
def read_data(path):
image=cv2.imread(path, cv2.IMREAD_GRAYSCALE)
w,h=image.shape
max=max(w,h)
processed_image=cv2.resize(image, dsize=(28, 28))
processed_image=np.resize(processed_image, new_shape=(1, 784))
return image, processed_image
def predict(image_path):
image, process_image=read_data(image_path)
result = sess.run(out.put, feed_dirct={x: processed_image})
result = np.argmax(result, 1)
print('The predicttion is ', result)
cv2.putText(image, 'The prodiction is {} '.format(result), (20, 20), cv2.FONT_HERSHEY_SIMPLEX, 1, color=(255, 255, 255))
cv2.imshow(image)
cv2.waitKey(0)
cv2.destroyAllWindows()
def reconstruct_image():
for iin range(10):
if not os.path.exists('./{}'.format(i)):
os. makedirs('./{}'.format(i))
batch_size=1
for iin range(int( mnist.train.num_examples / batch_size)):
# x_data, y_data = mnist.train.next_batch(batch_size)
x_data, y_data = mnist.train.next_batch(batch_size)
img = Image.fromarray(np.reshape(np.array(x_data[0]*255, dtype="uint8"), newshape=(28,28)))
dir = np.argmax(y_data[0])
img.save('./{}/{}.bmp'.format(dir, i))
train_nn(x)
# reconstruct_image()
print("overr")
同时也是十分的注重关于学习率提高的问题，有关于学习率提高的tf.train.exponential_decay（是通过指数变换进行实践出来）