神经网络编程 | 简单三层神经网络代码

__init __()：初始化函数，需要指定一个学习率。

train(): 训练函数。

query() 函数接受神经网络的输入，返回网络的输出。

神经网络代码

class neuralNetwork:
    
    def __init__(self, inputnodes, hiddennodes, outputnodes, learningrate):

        self.inodes = inputnodes
        self.hnodes = hiddennodes
        self.onodes = outputnodes

        self.wih = numpy.random.normal(0.0, pow(self.inodes, -0.5), (self.hnodes, self.inodes))
        self.who = numpy.random.normal(0.0, pow(self.hnodes, -0.5), (self.onodes, self.hnodes))

        self.lr = learningrate

        self.activation_function = lambda x: scipy.special.expit(x)
        
        pass

    
    def train(self, inputs_list, targets_list):

        inputs = numpy.array(inputs_list, ndmin=2).T
        targets = numpy.array(targets_list, ndmin=2).T
        
        hidden_inputs = numpy.dot(self.wih, inputs)
        hidden_outputs = self.activation_function(hidden_inputs)
        
        final_inputs = numpy.dot(self.who, hidden_outputs)
        final_outputs = self.activation_function(final_inputs)

        output_errors = targets - final_outputs
        hidden_errors = numpy.dot(self.who.T, output_errors) 
        
        self.who += self.lr * numpy.dot((output_errors * final_outputs * (1.0 - final_outputs)), numpy.transpose(hidden_outputs))
        
        self.wih += self.lr * numpy.dot((hidden_errors * hidden_outputs * (1.0 - hidden_outputs)), numpy.transpose(inputs))
        
        pass

    def query(self, inputs_list):

        inputs = numpy.array(inputs_list, ndmin=2).T
        
        hidden_inputs = numpy.dot(self.wih, inputs)
        hidden_outputs = self.activation_function(hidden_inputs)
        
        final_inputs = numpy.dot(self.who, hidden_outputs)
        final_outputs = self.activation_function(final_inputs)
        
        return final_outputs

传参

使用比输入节点的数量小的值，强制网络尝试总结输入的主要特点。如果选择太少的隐藏层节点，会限制网络的能力，使网络难以找到足够的特征或模式

应该选择多少个隐藏层节点，并不存在一个最佳方法

大的学习率会导致在梯度下降过程中有一些来回跳动和超调，减小学习率可以延长学习时间

input_nodes = 784
hidden_nodes = 200
output_nodes = 10

learning_rate = 0.1

n = neuralNetwork(input_nodes,hidden_nodes,output_nodes, learning_rate)

导入mnist训练数据集

from google.colab import drive
drive.mount('./mount')

training_data_file = open("mount/My Drive/Colab Notebooks/mnist_data/mnist_train.csv",'r')
training_data_list = training_data_file.readlines()
training_data_file.close()

训练

选择0.01作为范围最低点，避免0值输入会人为地造成权重更新失败。没有选择 0.99作为输入的上限值，是因为不需要避免输入1.0会造成这个问题。只需要避免输出值为1.0。

epochs = 10

for e in range(epochs):
    for record in training_data_list:
        all_values = record.split(',')
        inputs = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
        targets = numpy.zeros(output_nodes) + 0.01
        targets[int(all_values[0])] = 0.99
        n.train(inputs, targets)
        pass
    pass

测试集测试

test_data_file = open("mount/My Drive/Colab Notebooks/mnist_data/mnist_test.csv", 'r')
test_data_list = test_data_file.readlines()
test_data_file.close()

scorecard = []

for record in test_data_list:

    all_values = record.split(',')
    correct_label = int(all_values[0])
    inputs = (numpy.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
    outputs = n.query(inputs)
    label = numpy.argmax(outputs)

    if (label == correct_label):
        scorecard.append(1)
    else:
        scorecard.append(0)
        pass
    
    pass


scorecard_array = numpy.asarray(scorecard)
print ("performance = ", scorecard_array.sum() / scorecard_array.size)


#performance =  0.9716

尝试识别自己写的手写数字

修改图片尺寸

import glob
from PIL import Image
import numpy
import os

srcPath = 'mount/My Drive/Colab Notebooks/ps_predata/'
name = os.listdir(srcPath)
i = 0

for img_file_name in glob.glob('mount/My Drive/Colab Notebooks/ps_predata/*.PNG'):
  image = Image.open(img_file_name)
  resized_image = image.resize((28,28),Image.ANTIALIAS)
  resized_image.show()
  resized_image_name = name[i]
  i=i+1
  resized_image.save('mount/My Drive/Colab Notebooks/ps_data/'+resized_image_name)

使用自己做的数据集进行测试

import imageio

our_own_dataset = []
k=0
count=0
for image_file_name in glob.glob("mount/My Drive/Colab Notebooks/ps_data/*.PNG"):
  count=count+1
  correct_label = int(image_file_name[-5:-4])
  img_array = imageio.imread(image_file_name, as_gray=True)
  img_data  = 255.0 - img_array.reshape(784)
  img_data = (img_data / 255.0 * 0.99) + 0.01
  outputs = n.query(img_data)
  label = numpy.argmax(outputs)
  #print(correct_label,label)
  if(correct_label==label):
    k=k+1
#print('correct',k)
#print('sum',count)
print('accuracy',k/count)

数据预处理遇到的问题：

像素值：mnist黑底白字，自己做的数据集白底黑字，灰度值取反

背景颜色：除了数字外的灰度值即背景应该统一，否则会成为干扰的“噪声”

数字颜色：数字灰度值太小有一定的影响

mnist手写字体数据集下载地址

https://pjreddie.com/media/files/mnist_test.csv

mnist_train.csv

Colab笔记本

https://colab.research.google.com/?pli=1#scrollTo=Nma_JWh-W-IF