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
from sklearn.preprocessing import MinMaxScaler from sklearn.preprocessing import OneHotEncoder scaler = MinMaxScaler() # 归一化 X_data = scaler.fit_transform(X_data) print('MinMaxScaler_trans_X_data') print(X_data) Y = OneHotEncoder().fit_transform(Y_data).todense() # one-hot编码 print('ne-hot_Y') print(Y) X=X_data.reshape(-1,8,8,1) # 转换为图片格式 print(X) # 划分训练集测试集 from sklearn.model_selection import train_test_split X_train,X_test,Y_train,Y_test = train_test_split(X,Y,test_size=0.2,random_state=0,stratify=Y) print(X_train.shape,X_test.shape,Y_train.shape,Y_test.shape)
MinMaxScaler_trans_X_data
One-hot_Y
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。
依据：四层卷积和三个最大池化层，Dropout层防止过拟合。
4.模型训练
from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense,Dropout,Flatten,Conv2D,MaxPool2D model = Sequential() # 建立模型 ks = (3, 3) # 卷积核的大小 input_shape = X_train.shape[1:] model.add(Conv2D(filters=16, kernel_size=ks, padding='same', input_shape=input_shape, activation='relu'))#第一层输入数据的shape要指定外，其他层的数据的shape框架会自动推导 model.add(MaxPool2D(pool_size=(2, 2)))# 池化层1 model.add(Dropout(0.25))# 防止过拟合，随机丢掉连接 model.add(Conv2D(filters=32, kernel_size=ks, padding='same', activation='relu'))# 二层卷积 model.add(MaxPool2D(pool_size=(2, 2)))# 池化层2 model.add(Dropout(0.25)) model.add(Conv2D(filters=64, kernel_size=ks, padding='same', activation='relu'))# 三层卷积 model.add(Conv2D(filters=128, kernel_size=ks, padding='same', activation='relu'))# 四层卷积 model.add(MaxPool2D(pool_size=(2, 2)))# 池化层3 model.add(Dropout(0.25)) model.add(Flatten())# 平坦层 model.add(Dense(128, activation='relu'))# 全连接层 model.add(Dropout(0.25)) model.add(Dense(10, activation='softmax'))# 激活函数softmax model.summary() # 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) score = model.evaluate(X_test,Y_test) score
import matplotlib.pyplot as plt def show_train_history(train_history, train, validation): plt.plot(train_history.history[train]) plt.plot(train_history.history[validation]) plt.title('Train History') plt.ylabel('train') plt.xlabel('epoch') plt.legend(['train', 'validation'], loc='upper left') plt.show() p = plt.figure(figsize=(15, 15)) a1 = p.add_subplot(2, 1, 1) show_train_history(train_history, 'accuracy', 'val_accuracy') # 准确率 plt.show() a2 = p.add_subplot(2, 1, 2) show_train_history(train_history, 'loss', 'val_loss') # 缺失率 plt.show()
准确率：
缺失率：
5.模型评价
- model.evaluate()
- 交叉表与交叉矩阵
- pandas.crosstab
- seaborn.heatmap
import seaborn as sns score = model.evaluate(X_test, y_test) print('score', score) y_pred = model.predict_classes(X_test) print('y_pred', y_pred[:10]) y_test1 = np.argmax(y_test, axis=1).reshape(-1) # 交叉表、交叉矩阵 y_true = np.array(y_test1)[0] pd.crosstab(y_true, y_pred, rownames=['true'], colnames=['predict']) # 交叉表对比预测数据和原数据 y_test1 = y_test1.tolist()[0] # 交叉矩阵 a = pd.crosstab(np.array(y_test1), y_pred, rownames=['Lables'], colnames=['Predict']) df = pd.DataFrame(a) # 转为dataframe sns.heatmap(df, annot=True, cmap="Reds", linewidths=0.2, linecolor='G') plt.show()
