def test_sum(self):
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
assert_array_equal(sum(x).data, x.data.sum())
assert_array_equal(sum(x, axis=0).data, x.data.sum(axis=0))
assert_array_equal(
sum(x, keepdims=True).data, x.data.sum(keepdims=True))
y = 2 * sum(x, axis=0)
y.backward()
assert_array_equal(x.grad.data, np.array([[2, 2, 2], [2, 2, 2]]))
def test_shape_check(self):
x = Variable(np.random.randn(1, 10))
b = Variable(np.random.randn(10))
y = x + b
loss = F.sum(y)
loss.backward()
self.assertEqual(b.grad.shape, b.shape)
def __call__(self, x, C=1.0, k=1):
"""Call loss function of VAE.
The loss value is equal to ELBO (Evidence Lower Bound)
multiplied by -1.
Args:
x (Variable or ndarray): Input variable.
C (int): Usually this is 1.0. Can be changed to control the
second term of ELBO bound, which works as regularization.
k (int): Number of Monte Carlo samples used in encoded vector.
"""
z_mean, z_log_var = self.encoder(x)
rec_loss = 0
for l in range(k):
z = self.encoder.sampling(z_mean, z_log_var)
y = self.decoder(z)
rec_loss += F.binary_cross_entropy(F.flatten(y), F.flatten(x)) / k
kl_loss = C * (z_mean ** 2 + F.exp(z_log_var) - z_log_var - 1) * 0.5
kl_loss = F.sum(kl_loss) / len(x)
return rec_loss + kl_loss
# coding: utf-8
if '__file__' in globals():
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([[1, 2, 3], [4, 5, 6]]))
c = Variable(np.array([[10, 20, 30], [40, 50, 60]]))
t = x + c
print(t)
y = F.sum(t, axis=0)
y.backward()
print(y)
print(x.grad)
x = Variable(np.random.randn(2, 3, 4, 5))
y = x.sum(keepdims=True)
print(y.shape)
def test_backward4(self):
x_data = np.random.rand(10, 20, 20)
f = lambda x: F.sum(x, axis=None)
self.assertTrue(gradient_check(f, x_data))
def test_datatype(self):
x = Variable(np.random.rand(10))
y = F.sum(x)
# np.float64ではなく0次元のnp.ndarrayを返す
self.assertFalse(np.isscalar(y))
if '__file__' in globals():
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([1, 2, 3, 4, 5, 6]))
y = F.sum(x)
y.backward()
print(y)
print(x.grad)
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
y = F.sum(x)
y.backward()
print(y)
print(x.grad)
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
y = F.sum(x, axis=0)
y.backward()
print(y)
print(x.grad)
x = Variable(np.random.randn(2, 3, 4, 5))
y = x.sum(keepdims=True)
print(y.shape)
def total_varitaion_loss(x):
a = (x[:, :, :-1, :-1] - x[:, :, 1:, :-1])**2
b = (x[:, :, :-1, :-1] - x[:, :, :-1, 1:])**2
return F.sum(a + b)
import numpy as np from dezero import Variable import dezero.functions as F x = Variable(np.array([1, 2, 3, 4, 5, 6])) y = F.sum(x) y.backward() print(y) print(x.grad) x = Variable(np.array([[1, 2, 3], [4, 5, 6]])) y = F.sum(x) y.backward() print(y) print(x.grad) x = Variable(np.array([[1, 2, 3], [4, 5, 6]])) y = F.sum(x, axis=0) y.backward() print(y) print(x.grad) x = Variable(np.random.randn(2, 3, 4, 5)) y = F.sum(x, keepdims=True) print(y.shape)
def mean_squared_error(x0, x1):
diff = x0 - x1
return F.sum(diff**2) / len(diff)
def test_forward(self):
x = Variable(np.arange(6).reshape(2, 3))
y = sum(x)
self.assertEqual(y.data, np.array(15))
def mean_squared_error(x0, x1): # より良いversionはFの中に実装されている(メモリ節約) diff = x0 - x1 return F.sum(diff ** 2) / len(diff)
def softmax1d(x):
y = F.exp(x)
sum_y = F.sum(y)
return y / sum_y
def test_backward5(self):
x_data = np.random.rand(10, 20, 20) * 100
f = lambda x: F.sum(x, axis=None, keepdims=True)
self.assertTrue(gradient_check(f, x_data))
if '__file__' in globals():
import os, sys
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
import numpy as np
from dezero import Variable
from dezero.utils import plot_dot_graph
import dezero.functions as F
x = Variable(np.array([1, 2, 3, 4, 5, 6]))
y = F.sum(x)
y.backward()
print(y)
print(x.grad)
def test_forward1(self):
x = Variable(np.random.rand(10))
y = F.sum(x)
expected = np.sum(x.data)
self.assertTrue(np.allclose(y.data, expected))
def test_backward3(self):
x_data = np.random.rand(10, 20, 20)
f = lambda x: F.sum(x, axis=2)
self.assertTrue(check_backward(f, x_data))
def test_backward(self):
x = Variable(np.arange(6).reshape(2, 3))
y = sum(x)
y.backward()
assert_equal(x.grad.data, np.ones_like(x.data))
if "__file__" in globals():
import os, sys
sys.path.append(os.path.join(os.path.dirname(__file__), ".."))
import numpy as np
import dezero.functions as F
from dezero import Variable
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
c = Variable(np.array([[10, 20, 30], [40, 50, 60]]))
t = x + c
y = F.sum(t) # 오류 발생. step 39에서 구현할 함수임
y = backward(retain_grad=True)
print(y.grad)
print(t.grad)
print(x.grad)
print(c.grad)
def test_forward2(self):
x = Variable(np.random.rand(10, 20, 30))
y = F.sum(x, axis=1)
expected = np.sum(x.data, axis=1)
self.assertTrue(np.allclose(y.data, expected))
def test_forward1(self):
x = Variable(np.array(2.0))
y = F.sum(x)
expected = np.sum(x.data)
self.assertTrue(array_allclose(y.data, expected))
def mean_squared_error(y1, y2):
N = y1.shape[0]
diff = y1 - y2
loss = F.sum(diff * diff) / N
return loss
def test_forward3(self):
x = Variable(np.random.rand(10, 20, 30))
y = F.sum(x, axis=1, keepdims=True)
expected = np.sum(x.data, axis=1, keepdims=True)
self.assertTrue(array_allclose(y.data, expected))
def softmax1d(x):
x = as_variable(x)
y = F.exp(x)
sum_y = F.sum(y)
return y / sum_y
def test_backward1(self):
x_data = np.random.rand(10)
f = lambda x: F.sum(x)
self.assertTrue(gradient_check(f, x_data))
if '__file__' in globals():
import os
import sys
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
import numpy as np
from dezero.utils import plot_dot_graph
from dezero import Variable
import dezero.functions as F
x = Variable(np.array([[1, 2, 3], [4, 5, 6]]))
c = Variable(np.array([[10, 20, 30], [40, 50, 60]]))
t = x + c
y = F.sum(t)
y.backward(retain_grad=True)
print(y.grad)
print(t.grad)
print(x.grad)
print(c.grad)
import numpy as np from dezero import Variable import dezero.functions as F x0 = Variable(np.array([1, 2, 3])) x1 = Variable(np.array([10])) z = x0 + x1 # y = z.sum() y = F.sum(z) print(y) y.backward() print(x1.grad)
