import load_data [tr_data, va_data, te_data] = load_data.load_data_wrapper() print(list(tr_data)[0][1]) print(list(va_data)[0][1]) print(list(te_data)[0][1])
buffer = []
# extract and cluster centers as list of convolutional kernles
kernels = [self.hlf((patch.reshape(patch_size)-np.min(patch))/(np.max(patch) - np.min(patch)),0.5) for patch in kmeans.cluster_centers_]
return kernels
def hlf(self, in_array, thr=0.0):
array_out = in_array.copy()
array_out[array_out > thr] = 1
array_out[array_out <= thr] = -1
return array_out
if __name__ is '__main__':
'''
make figures for convolutional layre
'''
__, patches_data, __, test_data = load_data.load_data_wrapper()
#%%
np.random.seed(seed=0)
test_data = [(load_data.random_maniputlate_image(img.reshape(28,28)), key) for img, key in test_data[:49]]
t0 = time.time()
cnn = Convolution_Layer(patches_data, 25, (8,8))
dt = time.time() - t0
plt.figure(figsize=(4.2, 4.5))
for i, patch in enumerate(cnn.kernels):
plt.subplot(int(np.sqrt(cnn.n_clusters)), int(np.sqrt(cnn.n_clusters)), i + 1)
plt.imshow(patch, cmap=plt.cm.gray,
# coding=gbk
import load_data
import network
# 读取训练、测试和验证数据
training_data, validation_data, test_data = load_data.load_data_wrapper()
net = network.Network([784, 30, 10], cost=network.CrossEntropyCost)
net.large_weight_initializer()
x = net.SGD(training_data, 30, 10, 0.5,
evaluation_data=test_data, monitor_evaluation_accuracy=True)
print(x)
#!/usr/bin/python3
import network
import pickle
import load_data
if __name__ == "__main__":
nw = network.Network()
model = "model/1.model.pkl"
f = open(model, "rb")
nw.bias = pickle.load(f)
nw.weight = pickle.load(f)
nw.sizes = pickle.load(f)
_, _, te_data = load_data.load_data_wrapper()
td = list(te_data)
correct = nw.evaluate(td)
print("res: %d / %d" % (correct, len(td)))