# Liniarize II
nn = FeedForward([
LinearizeLayer(32, 32, 3),
FullyConnected(32 * 32 * 3, 300, identity),
Tanh(),
FullyConnected(300, 200, identity),
Tanh(),
FullyConnected(200, 10, identity),
SoftMax()
])
# # Convolutional I
nn = FeedForward([
ConvolutionalLayer(3, 32, 32, 6, 5, 1),
MaxPoolingLayer(2),
ReluLayer(),
ConvolutionalLayer(6, 14, 14, 16, 5, 1),
MaxPoolingLayer(2),
ReluLayer(),
LinearizeLayer(16, 5, 5),
FullyConnected(400, 300, relu),
FullyConnected(300, 10, relu),
SoftMax()
])
# Convolutional II
# nn = FeedForward([
# ConvolutionalLayer(3, 32, 32, 6, 5, 1),
# ReluLayer(),
# ConvolutionalLayer(6, 28, 28, 16, 5, 1),
def setUp(self):
self.layer = MaxPoolingLayer((1, 2))
def __init__(self):
# Lenet
# input: 28x28
# conv1: (5x5x6)@s1p2 -> 28x28x6 {(28-5+2x2)/1+1}
# maxpool2: (2x2)@s2 -> 14x14x6 {(28-2)/2+1}
# conv3: (5x5x16)@s1p0 -> 10x10x16 {(14-5)/1+1}
# maxpool4: (2x2)@s2 -> 5x5x16 {(10-2)/2+1}
# conv5: (5x5x120)@s1p0 -> 1x1x120 {(5-5)/1+1}
# fc6: 120 -> 84
# fc7: 84 -> 10
# softmax: 10 -> 10
lr = 0.01
self.layers = []
self.layers.append(
ConvolutionLayer(inputs_channel=1,
num_filters=6,
width=5,
height=5,
padding=2,
stride=1,
learning_rate=lr,
name='conv1'))
self.layers.append(ReLu())
self.layers.append(
MaxPoolingLayer(width=2, height=2, stride=2, name='maxpool2'))
self.layers.append(
ConvolutionLayer(inputs_channel=6,
num_filters=16,
width=5,
height=5,
padding=0,
stride=1,
learning_rate=lr,
name='conv3'))
self.layers.append(ReLu())
self.layers.append(
MaxPoolingLayer(width=2, height=2, stride=2, name='maxpool4'))
self.layers.append(
ConvolutionLayer(inputs_channel=16,
num_filters=120,
width=5,
height=5,
padding=0,
stride=1,
learning_rate=lr,
name='conv5'))
self.layers.append(ReLu())
self.layers.append(Flatten())
self.layers.append(
FullyConnectedLayer(num_inputs=120,
num_outputs=84,
learning_rate=lr,
name='fc6'))
self.layers.append(ReLu())
self.layers.append(
FullyConnectedLayer(num_inputs=84,
num_outputs=10,
learning_rate=lr,
name='fc7'))
self.layers.append(Softmax())
self.lay_num = len(self.layers)
class TestMaxPoolingLayer(unittest.TestCase):
def setUp(self):
self.layer = MaxPoolingLayer((1, 2))
def test_forward_prop(self):
self.layer.set_input_shape((4, 4))
input = np.array([[-1, -2, 3, 4],
[5, 6, -7, -8],
[9, -10, 11, -12],
[-13, 14, -15, 16]], dtype=np.float64)
expected_output = np.array([[-1, 4],
[6, -7],
[9, 11],
[14, 16]], dtype=np.float64)
output = self.layer.forward_prop(input)
numpy.testing.assert_array_equal(output, expected_output)
def test_back_prop(self):
self.layer.set_input_shape((4, 16))
input = np.array([[1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2],
[3, 4, -3, -4, 3, 4, -3, -4, 3, 4, -3, -4, 3, 4, -3, -4],
[-3, -4, 3, 4, -3, -4, 3, 4, -3, -4, 3, 4, -3, -4, 3, 4],
[1, 2, -1, -2, 1, 2, -1, -2, 1, 2, -1, -2, 1, 2, -1, -2]],
dtype=np.float64)
self.layer.forward_prop(input)
out_grad = np.array([[1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1]])
expected_in_grad = np.array([[0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1],
[0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0],
[1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1],
[0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0]],
dtype=np.float64)
in_grad = self.layer.back_prop(out_grad)
numpy.testing.assert_array_equal(in_grad, expected_in_grad)
def test_get_output_shape(self):
self.layer.set_input_shape((280, 72))
self.assertEqual(self.layer.get_output_shape(), (280, 36))