def test_maxpool_layer_1_forward(self):
print('\n==================================')
print(' Test maxpool layer forward ')
print('==================================')
np.random.seed(123)
in_dim, out_dim = 3, 3
kernel_size, stride = 3, 1
x = np.random.randn(1, in_dim, 5, 5)
maxpool_layer = MaxPoolingLayer(kernel_size, stride)
maxpool_out = maxpool_layer.forward(x)
correct_out = [[[[1.65143654, 1.49138963, 1.49138963],
[2.20593008, 2.20593008, 2.18678609],
[2.20593008, 2.20593008, 2.18678609]],
[[0.9071052 , 0.9071052 , 0.92746243],
[0.68822271, 0.68822271, 0.92746243],
[2.39236527, 2.39236527, 2.23814334]],
[[1.75488618, 1.75488618, 1.75488618],
[1.75488618, 1.75488618, 1.75488618],
[1.16220405, 1.16220405, 1.41729905]]]]
e = rel_error(correct_out, maxpool_out)
print('Relative difference:', e)
self.assertTrue(e <= 5e-8)
def test_maxpool_layer_2_backward(self):
print('\n==================================')
print(' Test maxpool layer backward ')
print('==================================')
np.random.seed(123)
in_dim, out_dim = 3, 3
kernel_size, stride = 3, 1
x = np.random.randn(1, in_dim, 5, 5)
maxpool_layer = MaxPoolingLayer(kernel_size, stride)
maxpool_out = maxpool_layer.forward(x)
d_prev = np.random.randn(*maxpool_out.shape)
dx = maxpool_layer.backward(d_prev, 0.01)
correct_dx = [[[[0., 0., 0., 0., 0.], [1.03972709, 0., 0., 0., 0.],
[0., 0., -0.52939562, 0., 0.],
[0., -1.47139598, 2.06254556, 0., 0.],
[0., 0., 0., 0., 0.]],
[[0., -1.40066055, 0., 0., 0.],
[0., 0., 0., 0., -0.56802076],
[0., 0., 0.87459023, 0., 0.], [0., 0., 0., 0., 0.],
[0., -0.11714651, 0., 0., 0.18103513]],
[[0., 0., 0., 0., 0.], [0., 0., -0.19467321, 0., 0.],
[0., 0., 0., 0., -1.97788793], [0., 0., 0., 0., 0.],
[0., 0., 0.26287825, 0., 0.]]]]
dx_e = rel_error(correct_dx, dx)
print('Relative difference dx:', dx_e)
self.assertTrue(dx_e <= 5e-8)
def test_model(num_feat, num_classes):
classifier = CNN_Classifier()
classifier.add_layer(
'Conv-1', ConvolutionLayer(num_feat, 2, kernel_size=3, stride=1,
pad=1))
classifier.add_layer('ReLU', ReLU())
classifier.add_layer(
'Conv-2', ConvolutionLayer(2, 3, kernel_size=3, stride=1, pad=1))
classifier.add_layer('tanh', Tanh())
classifier.add_layer(
'Conv-3', ConvolutionLayer(3, 3, kernel_size=3, stride=1, pad=0))
classifier.add_layer('Sigmoid', Sigmoid())
classifier.add_layer('Max-pool - 1',
MaxPoolingLayer(kernel_size=2, stride=1))
classifier.add_layer('FC-4', FCLayer(12, num_classes))
classifier.add_layer('Softmax', SoftmaxLayer())
return classifier
# 아래 예시를 참고하여 과제에 주어진 대로 (혹은 과제와 별개로 원하는 대로) Layer를 추가하세요.
# Convolution Layer
CNN.add_layer('Conv Layer - 1', ConvolutionLayer(in_channels=in_channnel, out_channels=8, kernel_size=3, pad=1))
# ReLU Layer
CNN.add_layer('ReLU Layer - 1', ReLU())
# Convolution Layer
CNN.add_layer('Conv Layer - 2', ConvolutionLayer(in_channels=8, out_channels=8, kernel_size=3, pad=1))
# ReLU Layer
CNN.add_layer('ReLU Layer - 2', ReLU())
# Max-pool Layer
CNN.add_layer('Max-Pool Layer', MaxPoolingLayer(kernel_size=2, stride=2))
# FC Layer
CNN.add_layer('FC Example Layer - 1', FCLayer(input_dim=1568, output_dim=500))
# FC Layer
CNN.add_layer('FC Example Layer - 2', FCLayer(input_dim=500, output_dim=5))
# Softmax Layer
# 이 layer는 항상 마지막에 추가
CNN.add_layer('Softmax Layer', SoftmaxLayer())
# Model Architecture 출력
CNN.summary()
# Hyper-parameters
conv_dx = conv_layer.backward(conv_out)
conv_dW = conv_layer.dW
conv_db = conv_layer.db
print('Convolution Layer Backward Check')
print('dx Difference : ', (conv_dx - correct_conv_dx).sum())
print('dW Difference : ', (conv_dW - correct_conv_dW).sum())
print('db Difference : ', (conv_db - correct_conv_db).sum())
print()
# ===========================================================================
print(
'============================ 4. Pooling Layer ============================='
)
kernel_size = 2
max_pool = MaxPoolingLayer(kernel_size, stride=kernel_size)
x = np.array([[5, 1, 3, 10], [-1, 0, 4, -1], [-10, -1, 11, 1], [0, 1, -5, 3]],
dtype=np.float32).reshape(1, 1, 4, 4)
pooled = max_pool.forward(x)
d_pool = max_pool.backward(np.ones_like(pooled))
print('Difference pool out: ', (correct_pooled - pooled).sum())
print('Difference d_pool: ', (correct_d_pool - d_pool).sum())
print()
# ===========================================================================
print(
'=============================== 5. FC Layer ==============================='
)
fc_layer = FCLayer(2, 2)
# Add Layers, Layer 추가 예제
# 아래 예시를 참고하여 과제에 주어진 대로 (혹은 과제와 별개로 원하는 대로) Layer를 추가하세요.
# Convolution Layer
CNN.add_layer(
'Conv Example Layer',
ConvolutionLayer(in_channels=in_channnel,
out_channels=16,
kernel_size=5,
pad=1))
# ReLU Layer
CNN.add_layer('ReLU Example Layer', ReLU())
# Max-pool Layer
CNN.add_layer('Max-Pool Example Layer', MaxPoolingLayer(kernel_size=3,
stride=3))
# FC Layer
CNN.add_layer('FC Example Layer', FCLayer(input_dim=1234, output_dim=123))
# Softmax Layer
# 이 layer는 항상 마지막에 추가
CNN.add_layer('Softmax Layer', SoftmaxLayer())
# Model Architecture 출력
CNN.summary()
# Hyper-parameters
num_epochs = None
learning_rate = None
print_every = None
train_accuracy = []
valid_accuracy = []
best_epoch = -1
best_acc = -1
best_model = None
# =============================== EDIT HERE ===============================
# Add layers
CNN.add_layer('Conv-1', ConvolutionLayer(in_channels=in_channels, out_channels=4, kernel_size=3, pad=1))
CNN.add_layer('ReLU-1',ReLU())
CNN.add_layer('Conv-2', ConvolutionLayer(in_channels=4, out_channels=4, kernel_size=3, pad=1))
CNN.add_layer('ReLU-2',ReLU())
CNN.add_layer('Max-pool-1',MaxPoolingLayer(2,2))
CNN.add_layer('FC-1',FCLayer(784,500))
CNN.add_layer('ReLU-3',ReLU())
CNN.add_layer('FC-2',FCLayer(500,10))
CNN.add_layer('Softmax Layer',SoftmaxLayer())
# =========================================================================
CNN.summary()
print('Training Starts...')
num_batch = int(np.ceil(num_train / batch_size))
for epoch in range(1, num_epochs + 1):
start = time.time()
epoch_loss = 0.0