# MNIST 數據集的 Keras 序列模型示例 以下是構建簡單多層感知機（在第 5 章中詳細介紹）的一個小例子，用于對 MNIST 集中的手寫數字進行分類： ```py import keras from keras.datasets import mnist from keras.models import Sequential from keras.layers import Dense, Dropout from keras.optimizers import SGD from keras import utils import numpy as np # define some hyper parameters batch_size = 100 n_inputs = 784 n_classes = 10 n_epochs = 10 # get the data (x_train, y_train), (x_test, y_test) = mnist.load_data() # reshape the two dimensional 28 x 28 pixels # sized images into a single vector of 784 pixels x_train = x_train.reshape(60000, n_inputs) x_test = x_test.reshape(10000, n_inputs) # convert the input values to float32 x_train = x_train.astype(np.float32) x_test = x_test.astype(np.float32) # normalize the values of image vectors to fit under 1 x_train /= 255 x_test /= 255 # convert output data into one hot encoded format y_train = utils.to_categorical(y_train, n_classes) y_test = utils.to_categorical(y_test, n_classes) # build a sequential model model = Sequential() # the first layer has to specify the dimensions of the input vector model.add(Dense(units=128, activation='sigmoid', input_shape=(n_inputs,))) # add dropout layer for preventing overfitting model.add(Dropout(0.1)) model.add(Dense(units=128, activation='sigmoid')) model.add(Dropout(0.1)) # output layer can only have the neurons equal to the number of outputs model.add(Dense(units=n_classes, activation='softmax')) # print the summary of our model model.summary() # compile the model model.compile(loss='categorical_crossentropy', optimizer=SGD(), metrics=['accuracy']) # train the model model.fit(x_train, y_train, batch_size=batch_size, epochs=n_epochs) # evaluate the model and print the accuracy score scores = model.evaluate(x_test, y_test) print('\n loss:', scores[0]) print('\n accuracy:', scores[1]) ``` 我們從描述和訓練 Keras 模型得到以下輸出： ```py _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= dense_7 (Dense) (None, 128) 100480 _________________________________________________________________ dropout_5 (Dropout) (None, 128) 0 _________________________________________________________________ dense_8 (Dense) (None, 128) 16512 _________________________________________________________________ dropout_6 (Dropout) (None, 128) 0 _________________________________________________________________ dense_9 (Dense) (None, 10) 1290 ================================================================= Total params: 118,282 Trainable params: 118,282 Non-trainable params: 0 _________________________________________________________________ Epoch 1/10 60000/60000 [========================] - 3s - loss: 2.3018 - acc: 0.1312 Epoch 2/10 60000/60000 [========================] - 2s - loss: 2.2395 - acc: 0.1920 Epoch 3/10 60000/60000 [========================] - 2s - loss: 2.1539 - acc: 0.2843 Epoch 4/10 60000/60000 [========================] - 2s - loss: 2.0214 - acc: 0.3856 Epoch 5/10 60000/60000 [========================] - 3s - loss: 1.8269 - acc: 0.4739 Epoch 6/10 60000/60000 [========================] - 2s - loss: 1.5973 - acc: 0.5426 Epoch 7/10 60000/60000 [========================] - 2s - loss: 1.3846 - acc: 0.6028 Epoch 8/10 60000/60000 [========================] - 3s - loss: 1.2133 - acc: 0.6502 Epoch 9/10 60000/60000 [========================] - 3s - loss: 1.0821 - acc: 0.6842 Epoch 10/10 60000/60000 [========================] - 3s - loss: 0.9799 - acc: 0.7157 loss: 0.859834249687 accuracy: 0.788 ``` 您可以看到，在 Keras 中構建和訓練模型是多么容易。 您可以從他們記錄完備的網站 [https://keras.io](https://keras.io) 獲取有關 Keras 的更多信息。