15、手写数字识别-小数据集

1.手写数字数据集

• from sklearn.datasets import load_digits
• digits = load_digits()

2.图片数据预处理

• x：归一化MinMaxScaler()
• y：独热编码OneHotEncoder()或to_categorical
• 训练集测试集划分
• 张量结构

3.设计卷积神经网络结构

• 绘制模型结构图，并说明设计依据。

4.模型训练

• model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
• train_history = model.fit(x=X_train,y=y_train,validation_split=0.2, batch_size=300,epochs=10,verbose=2)

5.模型评价

• model.evaluate()
• 交叉表与交叉矩阵
• pandas.crosstab
• seaborn.heatmap

  1 from sklearn.datasets import load_digits
  2 from sklearn.model_selection import train_test_split
  3 from sklearn.preprocessing import MinMaxScaler, OneHotEncoder
  4 from tensorflow.keras.models import Sequential
  5 from tensorflow.keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D
  6 import matplotlib.pyplot as plt
  7 import numpy as np
  8 import pandas as pd
  9 import seaborn as sns
 10 
 11 
 12 def create_dataset():
 13     digits = load_digits()
 14     X_data = digits.data.astype(np.float32)
 15     Y_data = digits.target.astype(np.float32).reshape(-1, 1)  # 将Y_data变为一列
 16 
 17     return X_data, Y_data
 18 
 19 
 20 def process_data(X_data, Y_data):
 21     # 将属性缩放到一个指定的最大和最小值（通常是1-0之间）
 22     scaler = MinMaxScaler()
 23     X_data = scaler.fit_transform(X_data)
 24     print("MinMaxScaler_trans_X_data:")
 25     print(X_data)
 26     Y = OneHotEncoder().fit_transform(Y_data).todense()  # 进行oe-hot编码
 27     print("one-hot_Y:")
 28     print(Y)
 29     return X_data, Y
 30 
 31 
 32 def split_dataset(X_data, Y):
 33     # 转换为图片的格式（batch, height, width, channels）
 34     X = X_data.reshape(-1, 8, 8, 1)
 35     X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.2, random_state=0, stratify=Y)
 36     print(X_train.shape, X_test.shape, y_train.shape, y_test.shape)
 37     return X_train, X_test, y_train, y_test
 38 
 39 
 40 def digits_model(X_train):
 41     """ 构建模型 """
 42     model = Sequential()
 43 
 44     ks = (3, 3)
 45     input_shape = X_train.shape[1:]
 46 
 47     # 一层卷积
 48     model.add(Conv2D(filters=64, kernel_size=ks, padding='same', input_shape=input_shape, activation='relu'))
 49     # 池化层1
 50     model.add(MaxPool2D(pool_size=(2, 2)))
 51     model.add(Dropout(0.2))
 52     # 二层卷积
 53     model.add(Conv2D(filters=64, kernel_size=ks, padding='same', activation='relu'))
 54     # 池化层2
 55     model.add(MaxPool2D(pool_size=(2, 2)))
 56     model.add(Dropout(0.2))
 57 
 58     model.add(Conv2D(filters=64, kernel_size=ks, padding='same', activation='relu'))
 59     model.add(Conv2D(filters=128, kernel_size=ks, padding='same', activation='relu'))
 60     model.add(MaxPool2D(pool_size=(2, 2)))
 61     model.add(Dropout(0.2))
 62 
 63     model.add(Flatten())  # 平坦层
 64     model.add(Dense(128, activation='relu'))  # 全连接层
 65     model.add(Dropout(0.2))
 66     model.add(Dense(10, activation='softmax'))  # 激活函数
 67 
 68     print(model.summary())
 69 
 70     return model
 71 
 72 
 73 def train_model(model, X_train, y_train):
 74     """ 训练模型 """
 75     model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
 76     train_history = model.fit(x=X_train, y=y_train, validation_split=0.2, batch_size=256, epochs=10, verbose=2)
 77     return train_history
 78 
 79 
 80 def score_model(model, X_test, y_test):
 81     """ 评估模型 """
 82     return print(model.evaluate(X_test, y_test)[1])
 83 
 84 
 85 def test_model(model, X_test):
 86     """ 测试模型 """
 87     y_pre = model.predict_classes(X_test)
 88     return y_pre
 89 
 90 
 91 def crossrtab_matrix(y_test, y_pre):
 92     """
 93     交叉表、交叉矩阵
 94     查看预测数据与原数据对比
 95     """
 96     y_test = np.argmax(y_test, axis=1).reshape(-1)
 97     y_test = np.array(y_test)[0]
 98     # print(y_test)
 99     # print(type(y_test))
100     # print('================')
101     # print(type(y_pre))
102     crosstab = pd.crosstab(y_test, y_pre, rownames=['labels'], colnames=['predict'])
103     matrix = pd.DataFrame(crosstab)
104     sns.heatmap(matrix, annot=True, cmap="RdPu", linewidths=0.2, linecolor='pink')
105     plt.show()
106 
107 
108 def show_train_history(train_history, train, validation):
109     plt.plot(train_history.history[train])
110     plt.plot(train_history.history[validation])
111     plt.title("Train History")
112     plt.ylabel("train")
113     plt.xlabel("epoch")
114     plt.legend(['train', 'validation'], loc='upper left')
115     plt.show()
116 
117 
118 if __name__ == '__main__':
119     X_data, Y_data = create_dataset()
120     X_data, Y = process_data(X_data, Y_data)
121     X_train, X_test, y_train, y_test = split_dataset(X_data, Y)
122     model = digits_model(X_train)
123     train_history = train_model(model, X_train, y_train)
124     score_model(model, X_test, y_test)
125     y_pre = test_model(model, X_test)
126     crossrtab_matrix(y_test, y_pre)
127     show_train_history(train_history, 'accuracy', 'val_accuracy')  # 准确率
128     show_train_history(train_history, 'loss', 'val_loss')  # 损失率

运行结果如下：

 

可视化展示数据训练参数最佳结果：

 

 　　由上图可知，当轮数epochs为10时，损失率以及准确率达到最佳。

交叉矩阵结果: