def main():
plt.ion()
# cnn = Cnn.load("saved/cnn_12-28-2019_19-36")
(X_train, y_train), (X_test, y_test) = mnist.load_data()
labels = np.zeros((y_train.shape[0], 10, 1))
for index, x in enumerate(y_train):
labels[index, x] = [1]
images = X_train.reshape(
(X_train.shape[0], 1, X_train.shape[1], X_train.shape[2])) / 255
test_images = X_test.reshape(
(X_test.shape[0], 1, X_test.shape[1], X_test.shape[2])) / 255
# cnn.add_layer(Conv2d(16, 5, 1))
# cnn.add_layer(MaxPooling((16, 24, 24), 2, 2))
# cnn.add_layer(Conv2d(32, 3, 16))
# cnn.add_layer(MaxPooling((32, 10, 10), 2, 2))
# cnn.add_layer(Flatten((32, 5, 5)))
# cnn.add_layer(FullyConnected(800, 10))
# cnn.add_layer(Softmax())
cnn.add_layer(Conv2d(16, 5, 1))
cnn.add_layer(MaxPooling((16, 24, 24), 2, 2))
cnn.add_layer(Conv2d(32, 3, 16))
cnn.add_layer(MaxPooling((32, 10, 10), 2, 2))
cnn.add_layer(Flatten((32, 5, 5)))
cnn.add_layer(FullyConnected(800, 128))
cnn.add_layer(FullyConnected(128, 10))
cnn.add_layer(Softmax())
# cnn.add_layer(Conv2d(16, 5, 1))
# cnn.add_layer(MaxPooling((16, 24, 24), 2, 2))
# cnn.add_layer(Conv2d(32, 5, 16))
# cnn.add_layer(MaxPooling((32, 8, 8), 2, 2))
# cnn.add_layer(Flatten((32, 4, 4)))
# cnn.add_layer(FullyConnected(512, 10))
# cnn.add_layer(Softmax())
# cnn.add_layer(Conv2d(32, 5, 1))
# cnn.add_layer(MaxPooling((32, 24, 24), 2, 2))
# cnn.add_layer(Conv2d(64, 5, 32))
# cnn.add_layer(MaxPooling((64, 8, 8), 2, 2))
# cnn.add_layer(Flatten((64, 4, 4)))
# cnn.add_layer(FullyConnected(1024, 10))
# cnn.add_layer(Softmax())
signal.signal(signal.SIGINT, keyboard_interrupt_handler)
print('Start learning')
cnn.learn(images, labels)
index = cnn.predict(test_images[0])
print(y_test[0] + ' ' + index)
def test_calculate_prev_layer_error_multi_feature(self):
layer = Conv2d(2, 2, 1, [1, -1, 1, 2, 1, 2, 1, 1])
theta = np.array((1., 0., 1., 0., 2., 0., 2., 0.)).reshape((2, 2, 2))
res = layer.calculate_prev_layer_error(theta)
exp = np.array([3, 3, 0, 6, 7, 0, 3, 4, 0]).reshape((1, 3, 3))
np.testing.assert_allclose(res, exp)
def test_calculate_prev_layer_error(self):
layer = Conv2d(1, 2, 1, [1, -1, 1, 2])
theta = np.array((0.5, 0.3, 0.2, 0.7)).reshape((1, 2, 2))
res = layer.calculate_prev_layer_error(theta)
exp = np.array([0.5, -0.2, -0.3, 0.7, 1.8, -0.1, 0.2, 1.1,
1.4]).reshape((1, 3, 3))
np.testing.assert_allclose(res, exp)
def setUp(self) -> None:
layers = []
for i in range(1, 4):
layer = Conv2d(1, 1, 1)
layer.forward = self.mock_method(
[np.array([i, i, i]),
np.array([i * 10, i * 10, i * 10])])
layer.back = self.mock_method([
np.array([-i, -i, -i]),
np.array([-i * 10, -i * 10, -i * 10])
])
layer.update_weights = MagicMock()
layers.append(layer)
self.loss_function = CrossEntropy()
self.loss_function.loss = self.mock_method([0.1, 0.2])
self.loss_function.delta = self.mock_method([[0.2, -0.3, 0.1],
[0.5, -0.6, 0.1]])
self.images = np.array([
[1, 2, 3],
[4, 5, 6],
])
self.labels = np.array([
[1, 0, 0],
[0, 0, 1],
])
self.layers = layers
def test_convolve_multi_channels(self):
layer = Conv2d(3, 1, 2, [0.5, 0.5, 2, 2, 1.5, 1.5], [1, 1, 1])
image = np.arange(1, 9).reshape((2, 2, 2))
res = layer.convolve(image)
exp = np.array(([[[3 + 1., 4. + 1], [5. + 1, 6. + 1]],
[[12. + 1, 16. + 1], [20. + 1, 24. + 1]],
[[9. + 1, 12. + 1], [15. + 1, 18. + 1]]]))
np.testing.assert_array_equal(res, exp)
def test_calculate_average_weights_derivative(self):
layer = Conv2d(1, 2, 1, [1, -1, 1, 2])
image = np.array(([1, 2, 1], [2, 3, 1], [2, 1, 1])).reshape((1, 3, 3))
theta = np.array(([2, 1, 1, 0])).reshape((1, 2, 2)).astype(np.float64)
cache_values = [
(image, theta),
]
res = layer.calculate_average_weights_derivative(cache_values)
exp = np.array(([[6., 8.], [9., 8.]])).reshape((1, 1, 2, 2))
np.testing.assert_array_equal(res, exp)
def test_calculate_average_weights_derivative_multiple_features(self):
layer = Conv2d(2, 2, 1, [1, 1, 1, 1, 1, 1, 1, 1])
image = np.array((1, 2, 1, 2, 3, 1, 2, 1, 1)).reshape((1, 3, 3))
theta = np.array(([
[[1, 0], [1, 0]],
[[0, 1], [0, 1]],
])).reshape((2, 2, 2))
cache_values = [(image, theta)]
res = layer.calculate_average_weights_derivative(cache_values)
exp = np.array([3., 5., 4., 4., 5., 2., 4., 2.]).reshape((2, 1, 2, 2))
np.testing.assert_allclose(res, exp)
def test_calculate_average_weights_derivative_multichannel(self):
layer = Conv2d(1, 2, 2, [1, 1, 1, 1, 1, 1, 1, 1])
image = np.array(
(1, 2, 1, 2, 3, 1, 2, 1, 1, 1, 2, 1, 0, 2, 2, 1, 1, 1)).reshape(
(2, 3, 3))
theta = np.array(([1, 0], [1, 0])).reshape((1, 2, 2))
cache_values = [
(image, theta),
]
res = layer.calculate_average_weights_derivative(cache_values)
exp = np.array([3., 5., 4., 4., 1., 4., 1., 3.]).reshape((1, 2, 2, 2))
np.testing.assert_allclose(res, exp)
def str2layer(directory, line):
switcher = {
"Conv2d":
lambda folder, args: Conv2d.load(folder, args),
"FullyConnected":
lambda folder, args: FullyConnected.load(folder, args),
"MaxPooling":
lambda folder, args: MaxPooling.load(folder, args),
"Flatten":
lambda folder, args: Flatten.load(folder, args),
"Softmax":
lambda folder, args: Softmax()
}
parts = line.rstrip().split(":")
func = switcher.get(parts[0])
return func(directory, parts[1])
def test_calculate_average_biases(self):
layer = Conv2d(2, 2, 1, [1, 1, 1, 1, 1, 1, 1, 1])
image1 = np.array((1, 2, 1, 2, 3, 1, 2, 1, 1)).reshape((1, 3, 3))
image2 = np.array((2, 1, 1, 2, 1, 1, 2, 2, 1)).reshape((1, 3, 3))
theta1 = np.array(([
[[1, 2], [1, 3]],
[[-1, 4], [2, 1]],
])).reshape((2, 2, 2))
theta2 = np.array(([
[[2, 1], [5, 1]],
[[1, 2], [5, 1]],
])).reshape((2, 2, 2))
cache_values = [(image1, theta1), (image2, theta2)]
res = layer.calculate_average_biases(cache_values)
exp_bias = np.array([8.0, 7.5]).reshape((2, 1))
np.testing.assert_allclose(res, exp_bias)
def test_calculate_average_weights_derivative_multiple_features_batch(
self):
layer = Conv2d(2, 2, 1, [1, 1, 1, 1, 1, 1, 1, 1])
image1 = np.array((1, 2, 1, 2, 3, 1, 2, 1, 1)).reshape((1, 3, 3))
image2 = np.array((2, 1, 1, 2, 1, 1, 2, 2, 1)).reshape((1, 3, 3))
theta1 = np.array(([
[[1, 0], [1, 0]],
[[0, 1], [0, 1]],
])).reshape((2, 2, 2))
theta2 = np.array(([
[[0, 1], [0, 1]],
[[1, 0], [1, 0]],
])).reshape((2, 2, 2))
cache_values = [(image1, theta1), (image2, theta2)]
res = layer.calculate_average_weights_derivative(cache_values)
exp = np.array([2.5, 3.5, 3.5, 3., 4.5, 2., 4., 2.5]).reshape(
(2, 1, 2, 2))
np.testing.assert_allclose(res, exp)
def test_init(self):
test = Conv2d(2, 3, 5)
self.assertEqual(test.kernel.shape, (2, 5, 3, 3))
def test_convolve(self):
layer = Conv2d(2, 2, 1, [1, -1, 1, 2, 1, 0, 0, 2], [2, 3])
image = np.arange(1, 10).reshape((1, 3, 3))
res = layer.convolve(image)
exp = np.array(([[[15., 18.], [24., 27.]], [[14., 17.], [23., 26.]]]))
np.testing.assert_array_equal(res, exp)
def test_img2vec(self):
layer = Conv2d(1, 2, 2)
image = np.arange(1, 33).reshape((2, 4, 4))
res = layer.img2vec(image)
def test_relu(self):
layer = Conv2d(1, 2, 1, [1, -1, 1, 2])
input = np.array(([-1, 3], [2, -1]))
exp = np.array(([0, 3], [2, 0]))
np.testing.assert_array_equal(layer.relu(input), exp)