# 输入数据为 mnist 数据集
x_test = x_test.astype('float32').reshape(-1, 28, 28, 1)
x_test = x_test / 255
y_test = keras.utils.to_categorical(y_test, 10)
score = model.evaluate(x_test, y_test)
return score[1]
if __name__ == '__main__':
model_path = '../ModelA_raw.hdf5'
model = load_model(model_path)
score = accuracy_mnist(model, mnist)
print('Origin Test accuracy: %.4f' % score)
acclst = []
bound_data_lst = get_bound_data_mnist(model, 10)
#print len(bound_data_lst)
index = np.random.choice(120, 30)
for i in range(30):
model_change = weight_shuffling(model,
Layer='dense_1',
neuron_index=index[i])
model_change.compile(loss='categorical_crossentropy',
optimizer='sgd',
metrics=['accuracy'])
#print 'Mutated Test accuracy: ',accuracy_cifar(model_change)
#acc= accuracy_in_bound_data_mnist(model_change,bound_data_lst)
#acclst.append(acc)
model_change.save('mutated/MODEL_A/ModelA_WShuff' + str(i) + '.hdf5')
#print 'Mutated accuracy in bound data(dense1):',[round(i,4) for i in acclst]
acclst = []
return : acc of mnist
'''
(x_train, y_train), (x_test, y_test) = mnist.load_data()
# 输入数据为 mnist 数据集
x_test = x_test.astype('float32').reshape(-1,28,28,1)
x_test = x_test / 255
y_test = keras.utils.to_categorical(y_test, 10)
score = model.evaluate(x_test, y_test)
return score[1]
if __name__=='__main__':
model_path='../ModelB_raw.hdf5'
model=load_model(model_path)
bound_data_lst = boundary.get_bound_data_mnist(model,10)#第一是第二的10倍以内，算边界值
unbound_data_lst = boundary.get_unbound_data_mnist(model,10)
path_lst=glob.glob('./mutated/MODEL_B/*.hdf5')
bound_acc_list=[]#变异模型在边界值集合上的准确率
unbound_acc_list=[]#变异模型在非边界值集合上的准确率
acc_list=[]
print('swj')
for index,path in enumerate(path_lst):
print index
mutant_model=load_model(path)
acc = accuracy_mnist(mutant_model,mnist)
acc_list.append(acc)
bound_acc =boundary.accuracy_in_bound_data_mnist(mutant_model,bound_data_lst)
unbound_acc =boundary.accuracy_in_unbound_data_mnist(mutant_model,unbound_data_lst)
bound_acc_list.append(bound_acc)
def train(filename='beta_mnist_shift_right3',
bound_ratio=2,
ramdon=True,
samplesize=100):
#(x_train, y_train), (__, __) = mnist.load_data()
npzfile = np.load('../mnist.npz')
x_train = npzfile['x_train']
y_train = npzfile['y_train']
#(x_train, y_train), (x_test, y_test) = mnist.load_data()
npzfile = np.load('../backup/' + filename + '.npz')
y_test = npzfile['y_test']
x_test = npzfile['x_test']
model_path = '../ModelA_raw.hdf5'
model = load_model(model_path)
len_bound = 0
#bound_ratio=2
if ramdon == False:
x_bound, y_bound = boundary.get_bound_data_mnist(
model, x_test, y_test, bound_ratio)
#x_train = x_bound
#y_train = y_bound
x_train = np.concatenate((x_bound, x_train), axis=0)
y_train = np.concatenate((y_bound, y_train), axis=0)
len_bound = len(x_bound)
else:
x_random, y_random = random_select(samplesize, x_test, y_test)
#x_train = x_random
#y_train = y_random
x_train = np.concatenate((x_random, x_train), axis=0)
y_train = np.concatenate((y_random, y_train), axis=0)
#x_train = x_bound
#y_train = y_bound
x_train = x_train.astype('float32').reshape(-1, 28, 28, 1)
x_test = x_test.astype('float32').reshape(-1, 28, 28, 1)
x_train = x_train / 255
x_test = x_test / 255
y_test = keras.utils.to_categorical(y_test, 10)
y_train = keras.utils.to_categorical(y_train, 10)
score = model.evaluate(x_test, y_test)
#print('Test Loss: %.4f' % score[0])
print('Test accuracy: %.4f' % score[1])
origin_acc = score[1]
'''
npzfile=np.load('./backup/beta_mnist_shift_up2.npz')
y_beta= npzfile['y_test']
x_beta= npzfile['x_test']
'''
sgd = optimizers.SGD(lr=0.01)
model.compile(loss='categorical_crossentropy',
optimizer=sgd,
metrics=['accuracy'])
model.fit(x_train,
y_train,
batch_size=100,
epochs=5,
verbose=1,
validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test)
print('Test accuracy: %.4f' % score[1])
return score[1], len_bound, origin_acc