def __call__(self, x):
x = F.relu(self.linear(x))
x = F.reshape(x, (-1,) + self.to_shape) # reshape to (-1, C, H, W)
x = F.relu(self.deconv(x))
x = self.conv(x)
x = F.sigmoid(x)
return x
def forward(self, x):
x = F.relu(self.conv1_1(x))
x = F.relu(self.conv1_2(x))
x = F.pooling(x, 2, 2)
x = F.relu(self.conv2_1(x))
x = F.relu(self.conv2_2(x))
x = F.pooling(x, 2, 2)
x = F.relu(self.conv3_1(x))
x = F.relu(self.conv3_2(x))
x = F.relu(self.conv3_3(x))
x = F.pooling(x, 2, 2)
x = F.relu(self.conv4_1(x))
x = F.relu(self.conv4_2(x))
x = F.relu(self.conv4_3(x))
x = F.pooling(x, 2, 2)
x = F.relu(self.conv5_1(x))
x = F.relu(self.conv5_2(x))
x = F.relu(self.conv5_3(x))
x = F.pooling(x, 2, 2)
x = F.reshape(x, (x.shape[0], -1))
x = F.dropout(F.relu(self.fc6(x)))
x = F.dropout(F.relu(self.fc7(x)))
x = self.fc8(x)
return x
def test_reshape(self):
x = Variable(np.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]))
y = reshape(x, (6, ))
y.backward()
assert_array_equal(y.data, np.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0]))
assert_array_equal(x.grad.data,
np.array([[1.0, 1.0, 1.0], [1.0, 1.0, 1.0]]))
def test_backward(self):
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
y = reshape(x, (6, ))
y.backward(retain_grad=True)
self.assertEqual(y.grad.shape, (6, ))
self.assertEqual(x.grad.shape, x.shape)
self.assertTrue((x.grad.data == np.ones((2, 3))).all())
def test_reshape(self):
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
y = F.reshape(x, (6, ))
y.backward(retain_grad=True)
self.assertEqual(y.shape, (6, ))
self.assertTrue(np.allclose(y.data, np.array([1, 2, 3, 4, 5, 6])))
self.assertTrue(
np.allclose(x.grad.data, np.array([[1, 1, 1], [1, 1, 1]])))
def forward(self, x):
x = F.relu(self.conv1(x)) # (OH, OW)=(28, 28)
x = F.pooling(x, 2, 2) # (OH, OW)=(14, 14)
#x = F.relu(self.conv2(x))
#x = F.pooling(x, 2, 2)
x = F.reshape(x, (x.shape[0], -1)) # (14, 14)->(196, )
x = F.dropout(F.relu(self.fc3(x)))
#x = F.dropout(F.relu(self.fc4(x)))
x = self.fc5(x)
return x
def forward(self, x):
x = F.relu(self.conv1_1(x)) # (OH, OW)=(28, 28)
x = F.relu(self.conv1_2(x)) # (OH, OW)=(28, 28)
x = F.pooling(x, 2, 2) # (OH, OW)=(14, 14)
x = F.relu(self.conv2_1(x)) # (OH, OW)=(14, 14)
x = F.relu(self.conv2_2(x)) # (OH, OW)=(14, 14)
x = F.pooling(x, 2, 2) # (OH, OW)=(7, 7)
x = F.reshape(x, (x.shape[0], -1)) # (7, 7)->(49, )
x = F.dropout(F.relu(self.fc3(x)))
x = self.fc4(x)
return x
import os, sys sys.path.append(os.path.join(os.path.dirname(__file__), '..')) import numpy as np from dezero import Variable import dezero.functions as F x = Variable(np.array([[0, 1, 2], [3, 4, 5]])) y = F.reshape(x, (6, )) # or y = x.reshape(6) y.backward(retain_grad=True) print(y.grad) x = Variable(np.array([[1, 2, 3], [4, 5, 6]])) y = F.transpose(x) # or y = x.T y.backward() print(x.grad)
import dezero.layers as L
from dezero import DataLoader
from dezero.models import Sequential
from dezero.optimizers import Adam
use_gpu = dezero.cuda.gpu_enable
max_epoch = 5
batch_size = 128
hidden_size = 62
fc_channel, fc_height, fc_width = 128, 7, 7
gen = Sequential(L.Linear(1024), L.BatchNorm(), F.relu,
L.Linear(fc_channel * fc_height * fc_width), L.BatchNorm(),
F.relu,
lambda x: F.reshape(x, (-1, fc_channel, fc_height, fc_width)),
L.Deconv2d(fc_channel // 2, kernel_size=4, stride=2, pad=1),
L.BatchNorm(), F.relu,
L.Deconv2d(1, kernel_size=4, stride=2, pad=1), F.sigmoid)
dis = Sequential(L.Conv2d(64, kernel_size=4, stride=2, pad=1), F.leaky_relu,
L.Conv2d(128, kernel_size=4, stride=2, pad=1),
L.BatchNorm(), F.leaky_relu, F.flatten, L.Linear(1024),
L.BatchNorm(), F.leaky_relu, L.Linear(1), F.sigmoid)
def init_weight(dis, gen, hidden_size):
# Input dummy data to initialize weights
batch_size = 1
z = np.random.rand(batch_size, hidden_size)
fake_images = gen(z)
def _global_average_pooling_2d(x):
N, C, H, W = x.shape
h = F.average_pooling(x, (H, W), stride=1)
h = F.reshape(h, (N, C))
return h
import os, sys sys.path.append(os.path.join(os.path.dirname(__file__), '..')) import numpy as np from dezero import Variable import dezero.functions as F x = Variable(np.array([[0, 1, 2], [3, 4, 5]])) y = F.reshape(x, (6, )) # y = x.reshape(6) y.backward(retain_grad=True) print(x.grad) x = Variable(np.array([[1, 2, 3], [4, 5, 6]])) y = F.transpose(x) # y = x.T y.backward() print(x.grad)
def test_forward(self):
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
y = reshape(x, (6, ))
self.assertEqual(y.shape, (6, ))
def test_reshape_backward(self):
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
y = F.reshape(x, (6, ))
y.backward(create_graph=False)
assert x.grad.shape == (2, 3)
def test_reshape_forward(self):
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
y = F.reshape(x, (6, ))
assert y.shape == (6, )
