def main():
layer1 = 28 * 28
layer2 = 100
layer3 = 10
learning_rate = 0.3
nn = net.NeuralNet(layer1, layer2, layer3, learning_rate)
nn.load_model()
score = []
util.for_each_record('dataset/mnist_test.csv', (
lambda label, pixels:
score.append(label == np.argmax(nn.query(pixels)))
))
score = np.asfarray(score)
print('\nScore =', score.sum() / score.size)
def main():
layer1 = 28 * 28
layer2 = 100
layer3 = 10
learning_rate = 0.1
nn = net.NeuralNet(layer1, layer2, layer3, learning_rate)
nn.load_model()
score = []
util.for_each_image_in_path(path, (
lambda label, pixels:
score.append(label == np.argmax(nn.query(pixels)))
))
score = np.asfarray(score)
print('\nScore =', score.sum() / score.size)
def main():
print('Test with Epoch 1~10')
layer1 = 28 * 28
layer2 = 100
layer3 = 10
learning_rate = 0.1
nn = net.NeuralNet(layer1, layer2, layer3, learning_rate)
for i in range(1, 11):
nn.train('dataset/mnist_train.csv', i)
score = []
util.for_each_record('dataset/mnist_test.csv',
(lambda label, pixels: score.append(
label == np.argmax(nn.query(pixels)))))
score = np.asfarray(score)
print(str(i) + '\t' + str(score.sum() / score.size))
def main():
layer1 = 28 * 28
layer2 = 100
layer3 = 10
learning_rate = 0.3
nn = net.NeuralNet(layer1, layer2, layer3, learning_rate)
nn.load_model()
images = []
for label in range(10):
target = np.zeros(layer3) + 0.01
target[label] = 0.99
image = nn.inverse(target)
print(image)
images.append(image)
fig = plt.figure()
max_index = len(images) - 1
def get_pixels(index):
if index >= max_index:
plt.close()
ret = images[index].reshape(28, 28)
print('index: ', index)
return ret
im = plt.imshow(get_pixels(0),
cmap='Greys',
interpolation='None',
animated=True)
def updatefig(frame, *args):
im.set_array(get_pixels(frame))
return im,
ani = animation.FuncAnimation(fig, updatefig, interval=4000, blit=True)
plt.show()