版权声明：本文为博主原创文章，转载请注明出处。 https://blog.csdn.net/u012319493/article/details/87125745
导入数据集
from keras.datasets import mnist
import numpy as np
(X_train, Y_train), (X_test, Y_test) = mnist.load_data()
print(np.shape(X_train), np.shape(X_test))
(60000, 28, 28) (10000, 28, 28)
数据预处理
# 把三维的张量（tensor）转换成二维的数值
X_train = X_train.reshape(60000, 28*28)
X_test = X_test.reshape(10000, 28*28)
print(np.shape(X_train), np.shape(X_test))
(60000, 784) (10000, 784)
# 每个图片的的数值位于 (0, 255) --> (0, 1)：可加速模型训练
X_train /= 255
X_test /= 255
# 设定每个数值是 float32 型的：GPU 可使用这些变量加速训练
X_train = X_train.astype("float32")
X_test = X_test.astype("float32")
# 1 of K encoding: [0, 1, 2], y=1->(0, 1, 0), y=2->(0, 0, 1), y=0->(1, 0, 0)
from keras.utils import np_utils
num_classes = 10
Y_train = np_utils.to_categorical(Y_train, num_classes)
Y_test = np_utils.to_categorical(Y_test, num_classes)
print(Y_train)
[[0. 0. 0. … 0. 0. 0.]
[1. 0. 0. … 0. 0. 0.]
[0. 0. 0. … 0. 0. 0.]
…
[0. 0. 0. … 0. 0. 0.]
[0. 0. 0. … 0. 0. 0.]
[0. 0. 0. … 0. 1. 0.]]
搭建深度学习神经网络
以下为所设计的神经网络：
# 利用 keras 定义深度学习网络
from keras.models import Sequential
from keras.layers import Dense
from keras.optimizers import SGD
model = Sequential()
# 添加了输入层的信息
model.add(Dense(100, activation='sigmoid', input_shape=(784,))) # batch_size = 20, 30, 50, ...
# 添加第一个隐含层的信息
model.add(Dense(100, activation='relu'))
# 添加第二个隐含层
model.add(Dense(num_classes, activation='softmax'))
# 目标函数，优化算法，评估方法（准确，AUC）
model.compile(loss='categorical_crossentropy', optimizer=SGD(), metrics=['accuracy'])
# 训练模型
model.fit(X_train, Y_train, batch_size=50, epochs=10, verbose=1, validation_data=(X_test, Y_test))
score = model.evaluate(X_test, Y_test, verbose=1)
print("Test loss: ", score[0]) # 目标函数，损失函数
print("Test accuracy: ", score[1]) # 正确率
结果：
Instructions for updating:
Use tf.cast instead.
Train on 60000 samples, validate on 10000 samples
Epoch 1/10
60000/60000 [==============================] - 5s 81us/step - loss: 0.7790 - acc: 0.7981 - val_loss: 0.4146 - val_acc: 0.8900
Epoch 2/10
60000/60000 [==============================] - 3s 52us/step - loss: 0.3786 - acc: 0.8926 - val_loss: 0.3383 - val_acc: 0.9064
Epoch 3/10
60000/60000 [==============================] - 3s 54us/step - loss: 0.3267 - acc: 0.9070 - val_loss: 0.3010 - val_acc: 0.9156
Epoch 4/10
60000/60000 [==============================] - 3s 52us/step - loss: 0.3021 - acc: 0.9120 - val_loss: 0.2824 - val_acc: 0.9176
Epoch 5/10
60000/60000 [==============================] - 3s 53us/step - loss: 0.2856 - acc: 0.9167 - val_loss: 0.2696 - val_acc: 0.9197
Epoch 6/10
60000/60000 [==============================] - 3s 52us/step - loss: 0.2687 - acc: 0.9206 - val_loss: 0.2563 - val_acc: 0.9250
Epoch 7/10
60000/60000 [==============================] - 3s 52us/step - loss: 0.2571 - acc: 0.9241 - val_loss: 0.2679 - val_acc: 0.9238
Epoch 8/10
60000/60000 [==============================] - 3s 52us/step - loss: 0.2458 - acc: 0.9269 - val_loss: 0.2433 - val_acc: 0.9270
Epoch 9/10
60000/60000 [==============================] - 3s 54us/step - loss: 0.2472 - acc: 0.9266 - val_loss: 0.2378 - val_acc: 0.9279
Epoch 10/10
60000/60000 [==============================] - 3s 53us/step - loss: 0.2280 - acc: 0.9322 - val_loss: 0.2294 - val_acc: 0.9320
10000/10000 [==============================] - 0s 32us/step
Test loss: 0.22936600434184073
Test accuracy: 0.932