# use dropout to improve neural network 
The from __future__ import print_function 
The import numpy as np 
The from keras. Datasets import mnist 
The from keras. Models import Sequential 
The from keras. The layers. The core import Dense, Activation 
The from keras. Optimizers import SGD 
The from keras. Utils import np_utils 
Np, the random seed (1671) # repetitive setting 

# network and training 
NB_EPOCH=250 
BATCH_SIZE=128 
VERBOSE=1 
NB_CLASSES=10 
The OPTIMIZER=SGD () 
N_HIDDEN=128 
VALIDATION_SPLIT=0.2 
DROPOUT=0.3 
# data: the mixed and divided into training set and testing set 
(X_train y_train), (X_test, y_test)=mnist. Load_data () 
28 * 28 # X_train is 60000 lines of data, can be transformed into 60000 * 784 
RESHAPED=784 
# 
X_train=X_train. Reshape (60000, RESHAPED) 
X_test=X_test. Reshape (10000, RESHAPED) 
X_train=X_train. Astype (' float32 ') 
X_test=X_test. Astype (' float32 ') 
# normalized 
X_train/=255 
X_test/=255 
# transform class vector into binary classification matrix 
Y_train=np_utils. To_categorical (Y_train NB_CLASSES) 
Y_test=np_utils. To_categorical (Y_test NB_CLASSES) 
# N_HIDDEN a hidden layer, 10 a output 
The model=Sequential () 
Model. The add (Dense (N_HIDDEN, input_shape=(RESHAPED,))) 
Model. The add (Activation (' relu)) 
Model. The add (Dropout (Dropout)) 
Model. The add (Dense (N_HIDDEN)) 
Model. The add (Activation (' relu)) 
Model. The add (Dropout (Dropout)) 
Model. The add (Dense (NB_CLASSES)) 
Model. The add (Activation (' softmax)) 
Model. The summary () 
Model.com from running (loss='categorical_crossentropy', 
Optimizer=optimizer, 
The metrics=[' accuracy ']) 
History=the model fit (X_train Y_train, 
Batch_size=batch_size, epochs=NB_EPOCH, 
Verbose=verbose, validation_split=validation_split) 
Score=model. The evaluate (X_test Y_test, verbose=verbose) 
Print (", "Test score, score [0]) 
Print (' Test accuracy: 'score [1])