深度学习01--Mnist手写体识别

纯记录自己在深度学习方面的学习历程--基于b站的人工智能许多up主的教学视频

1. 定义一个多层感知机的类

   import numpy as np
   from utils.features import prepare_for_training
   from utils.hypothesis import sigmoid, sigmoid_gradient
   
   # 定义一个类，叫多层感知机--multilayer perceptron
   class MultilayerPerceptron:
       # 初始化，data--数据,label--标签,layers--神经网络层数,normalize_data--归一化
       def __init__(self, data, labels, layers, normalize_data=False):
           data_processed = prepare_for_training(data, normalize_data=normalize_data)[0]
           self.data = data_processed
           self.labels = labels
           self.layers = layers  # 预计三层 28*28*1=784 25 10
           self.normalize_data = normalize_data
           self.ws = MultilayerPerceptron.w_init(layers)  # 对权重参数的初始化
   
       # 训练模块 max--迭代次数,alpha--学习率
       def train(self, max=1000, alpha=0.1):
           unroll_ws = MultilayerPerceptron.unroll(self.ws)
           (optimized_ws, loss_history) = MultilayerPerceptron.gradient_descent(self.data, self.labels, self.layers,
                                                                                unroll_ws, max, alpha)
           self.ws = MultilayerPerceptron.roll(optimized_ws, self.layers)
           return self.ws, loss_history
   
       # 测试
       def predict(self, data):
           data_processed = prepare_for_training(data, normalize_data=self.normalize_data)[0]
           num_examples = data_processed.shape[0]
           predictions = MultilayerPerceptron.forward(data_processed, self.layers, self.ws)
           return np.argmax(predictions, axis=1).reshape((num_examples, 1))
   
       # 对权重参数的初始化
       def w_init(layers):
           num_layers = len(layers)
           ws = {}
           # 实际循环两次，生成两个w矩阵：25*785 10*26
           for index in range(num_layers - 1):
               income = layers[index]
               outcome = layers[index + 1]
               # +1根据权重公式xw+b，考虑到b，例784*25+25=（784+1）*25
               ws[index] = np.random.rand(outcome, income + 1) * 0.05  # *0.05是使随机生成的值尽量小
           return ws
   
       # 将矩阵展开为一维向量
       def unroll(m