preface : This is based on tensorflow frame , Created with only one hidden layer BP neural network , Image recognition done , The content is also relatively simple , It's all my study notes .

–—-—-—-—-—-—-—-—-—-—-—-—–—-—-—-—-—-—-—-——-—-—-—-—-—-—-—-—-—-—-—-—-—-——-
–—-—-—-—-—-—-—-—-—-—-—-—–—-—-—-—-—-—-—-——-—-—-—-—-—-—-—-—-—-—-—-—-—-——-

1, load MNIST data set
from tensorflow.examples.tutorials.mnist import input_data mnist =
input_data.read_data_sets('MNIST_data', one_hot=True)
mnist Is a lightweight class . It uses Numpy Training is stored in the form of array , Checksum test data set , It's also Google Classic data set for image recognition .MNIST Dataset download link :
https://pan.baidu.com/s/1d9ty82 <https://pan.baidu.com/s/1d9ty82> password : jcam

2, function TensorFlow frame
import tensorflow as tf sess = tf.InteractiveSession() init = tf.global
_variables_initializer() sess.run(init)
TensorFlow The connection between the framework and the back end is called session, That is to say, we use session start-up TensorFlow frame ( I need to know more about it )

3, Predefined input values X, True value Y
X = tf.placeholder(tf.float32, shape=[None, 784]) Y = tf.placeholder(tf.float
32, shape=[None,10])
* X,Y Now represented by placeholders , You can TensorFlow When running a calculation , Calculate based on the specific value entered by the placeholder ;
* tf.float32 Is the type of storage ;shape=[None,
784] Is the data dimension size —— because MNIST The size of each picture in the dataset is 28*28 Of , When calculating, it will 28*28 2-D data is transformed into a 1-D , Count Reg 784 New vector of .None Indicates that its value size is variable , Means selected X,Y The number of
4, establish BP neural network
""" The way of generating with random sequence , Creating a neural network with a hidden layer .(784,300,10) """
#truncated_normal: Select the mean value of normal distribution =0.1 Nearby random value w1 = tf.Variable(tf.truncated_normal([784
,300],stddev=0.1)) w2 = tf.Variable(tf.zeros([300,10])) b1 =
tf.Variable(tf.zeros([300])) b2 = tf.Variable(tf.zeros([10]))
#relu,softmax Are activation functions L1 = tf.nn.relu(tf.matmul(X,w1)+b1) y =
tf.nn.softmax(tf.matmul(L1,w2)+b2)

BP The input layer of neural network has 784 Neurons , Hide layer 300 Neurons , Output layer 10 Neurons . Initialize weights at all levels w1,w2; Offset value of each stage b1,b2—— They are all generated by random sequence . Define hidden layers , Calculation method of output layer and its activation function .

5, Calculate error and optimize weight by gradient descent method
# Quadratic cost function : Calculate forecast y And true value Y Error between loss = tf.reduce_mean(tf.square(Y - y))
# Gradient descent method : Selection GradientDescentOptimizer optimizer , The learning rate is 0.5 train_step = tf.train
.GradientDescentOptimizer(0.5).minimize(loss)

Error is also called loss function , Cost function .TensorFlow There are a lot of built-in optimization algorithms in , Here we choose the simplest GradientDescentOptimizer Optimizer reduces cross entropy , Step is set as 0.5

6, Calculation accuracy
# Results are stored in a Boolean list correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(Y,1))
# Accuracy rate accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
*
tf.argmax() function : Is the index value corresponding to the maximum data value of the returned object in a certain dimension , Because the label vectors here are all generated by 0,1 form , So maximum 1 The index position is the corresponding category label
* tf.argmax(y,1) What is returned is for any input x Predicted label value ,tf.argmax(Y,1) Represents the correct label value
* correct_prediction
Here is the return of a Boolean array . In order to calculate the accuracy of our classification , We convert Boolean values to floating-point numbers to represent right and wrong , Then take the average . for example :[True, False, True,
True] Change to [1,0,1,1], The accuracy is 0.75
7, Other instructions
batch_xs,batch_ys = mnist.train.next_batch(batch_size) sess.run
(train_step,feed_dict=({X:batch_xs,Y:batch_ys})) acc = sess.run
(accuracy,feed_dict={X:mnist.test.images,Y:mnist.test.labels})
* batch_xs And batch_ys: From MNIST The number of batches in the data set : Data items and label items
* feed_dict=({X:batch_xs,Y:batch_ys} sentence : Yes batch_xs,batch_ys Value passed in for X,Y
Source code and effect display
# -*- coding:utf-8 -*- # -*- author:zzZ_CMing # -*- 2018/01/23;21:49 # -*-
python3.5 import tensorflow as tf from tensorflow.examples.tutorials.mnist
import input_data import os os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' # read MNIST data set
mnist = input_data.read_data_sets("MNIST_data",one_hot=True) # Set the size of each batch
batch_size =500 # How many batches are there ( Floor removal ) n_batch = mnist.train.num_examples//batch_size
# Predefined input values X, Output true value Y placeholder Is a placeholder X = tf.placeholder(tf.float32,[None,784]) Y =
tf.placeholder(tf.float32,[None,10]) """ The way of generating with random sequence , Creating a neural network with a hidden layer .(784,300,10)
""" #truncated_normal: Select the mean value of normal distribution =0.1 Nearby random value w1 =
tf.Variable(tf.truncated_normal([784,300],stddev=0.1)) w2 =
tf.Variable(tf.zeros([300,10])) b1 = tf.Variable(tf.zeros([300])) b2 =
tf.Variable(tf.zeros([10])) #relu,softmax Are activation functions L1 =
tf.nn.relu(tf.matmul(X,w1)+b1) y = tf.nn.softmax(tf.matmul(L1,w2)+b2)
# Quadratic cost function : Error between predicted value and real value loss = tf.reduce_mean(tf.square(Y - y))
# Gradient descent method : Selection GradientDescentOptimizer optimizer , The learning rate is 0.5 train_step =
tf.train.GradientDescentOptimizer(0.5).minimize(loss) # Results are stored in a Boolean list
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(Y,1)) # Accuracy rate
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))# initialize variable , activation tf frame
init = tf.global_variables_initializer() sess = tf.Session() sess.run(init)for i
in range(21): for batch in range(n_batch): batch_xs,batch_ys =
mnist.train.next_batch(batch_size)
sess.run(train_step,feed_dict=({X:batch_xs,Y:batch_ys})) acc =
sess.run(accuracy,feed_dict={X:mnist.test.images,Y:mnist.test.labels}) print(
"Iter " + str(i)+",Testing Accuracy "+str(acc))
Effect display :


*
iteration 10 The accuracy of the second time has arrived 90.85%, For today's technology , This accuracy is still relatively low .CNN Convolutional neural network has been able to achieve the accuracy of more complex image recognition 98% above , So I have to go further in my study
* The program runs slowly , Let's all learn
–—-—-—-—-—-—-—-—-—-—-—-—–—-—-—-—-—-—-—-——-—-—-—-—-—-—-—-—-—-—-—-—-—-——-
–—-—-—-—-—-—-—-—-—-—-—-—–—-—-—-—-—-—-—-——-—-—-—-—-—-—-—-—-—-—-—-—-—-——-

Series recommendations :

【 Supervised learning 】1:KNN Three methods of handwritten digit recognition based on Algorithm
<https://blog.csdn.net/zzz_cming/article/details/78938107>
–—-—-—-—-—-—-—-—-—-—-—-—–—-—-—-—-—-—-—-——-—-—-—-—-—-—-—-—-—-—-—-—-—-——-
【 Unsupervised learning 】1:K-means Introduction to algorithm principle , And code implementation
<https://blog.csdn.net/zzz_cming/article/details/79859490>
【 Unsupervised learning 】2:DBSCAN Introduction to algorithm principle , And code implementation
<https://blog.csdn.net/zzz_cming/article/details/79863036>
【 Unsupervised learning 】3:Density Peaks clustering algorithm ( Local density clustering )
<https://blog.csdn.net/zzz_cming/article/details/79889909>
–—-—-—-—-—-—-—-—-—-—-—-—–—-—-—-—-—-—-—-——-—-—-—-—-—-—-—-—-—-—-—-—-—-——-
【 Deep learning 】1: Perceptron principle , And multi-layer perceptron to solve XOR problem
<https://blog.csdn.net/zzz_cming/article/details/79031869>
【 Deep learning 】2:BP The principle of neural network , And resolution of exclusive or problems
<https://blog.csdn.net/zzz_cming/article/details/79118894>
【 Deep learning 】3:BP neural net recognition MNIST data set
<https://blog.csdn.net/zzz_cming/article/details/79136928>
【 Deep learning 】4:BP neural network +sklearn Digital recognition
<https://blog.csdn.net/zzz_cming/article/details/79182103>
【 Deep learning 】5:CNN Principle of convolutional neural network ,MNIST Dataset identification
<https://blog.csdn.net/zzz_cming/article/details/79192815>
【 Deep learning 】8:CNN Convolutional neural network recognition sklearn data set ( Source code attached )
<https://blog.csdn.net/zzz_cming/article/details/79691459>
【 Deep learning 】6:RNN Principle of recurrent neural network ,MNIST Dataset identification
<https://blog.csdn.net/zzz_cming/article/details/79235475>
【 Deep learning 】7:Hopfield neural network (DHNN) Principle introduction
<https://blog.csdn.net/zzz_cming/article/details/79289502>
–—-—-—-—-—-—-—-—-—-—-—-—–—-—-—-—-—-—-—-——-—-—-—-—-—-—-—-—-—-—-—-—-—-——-
TensorFlow Brief introduction to the framework <https://blog.csdn.net/zzz_cming/article/details/79235469>
–—-—-—-—-—-—-—-—-—-—-—-—–—-—-—-—-—-—-—-——-—-—-—-—-—-—-—-—-—-—-—-—-—-——-