def test_forward(self):
x = Variable(np.array(2.0))
y = fn.square(x)
expected = np.array(4.0)
self.assertEqual(
y.data, expected
) # 두 개체가 동일한지 여부를 판정 이외에도(assertGreater, self.assertTrue등)
def test_gradient_check(self):
x = Variable(np.random.rand(1))
y = fn.square(x)
y.backward()
num_grad = fn.numerical_diff(fn.square, x) # 수치미분
flg = np.allclose(x.grad, num_grad)
#np.allclose(a,b) : 기본값 |a-b|<=(atol + rtol * |b|) 만족하면 True (값이 얼마나 가까운지 판단)
self.assertTrue(flg)
def forward(self, x):
y = np.sin(x)
return y
def backward(self, gy):
x = self.inputs[0].data
gx = gy * np.cos(x)
return gx
def sin(x):
return Sin()(x)
def my_sin(x, threshold=0.0001):
y = 0
for i in range(100000):
c = (-1)**i / math.factorial(2 * i + 1)
t = c * x**(2 * i + 1)
y = y + t
if abs(t.data) < threshold:
break
return y
if __name__ == '__main__':
x = Variable(np.array(np.pi / 4))
y = my_sin(x)
y.backward()
plot_dot_graph(y, verbose=False, to_file='my_sin.png')
import os, sys; sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
import numpy as np
# Import core_simple explicitly
from dezero.core_simple import Variable
from dezero.core_simple import setup_variable
setup_variable()
def f(x):
y = x ** 4 - 2 * x ** 2
return y
def gx2(x):
return 12 * x ** 2 - 4
logs = []
x = Variable(np.array(2.0))
iters = 200
for i in range(iters):
print(i, x)
y = f(x)
x.cleargrad()
y.backward()
x.data -= 0.01 * x.grad
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
import numpy as np
from dezero.core_simple import Variable
from dezero.core_simple import setup_variable
setup_variable()
def sphere(x, y):
z = x**2 + y**2
return z
def metyas(x, y):
z = 0.26 * (x**2 + y**2) - 0.48 * x * y
return z
def goldstein(x, y):
z = (1 + (x + y + 1)**2 * (19 - 14 * x + 3 * x ** 2 - 14 * y + 6 * x * y + 3 * y ** 2)) * \
(30 + (2 * x - 3 * y)**2 * (18 - 32 * x + 12 *
x ** 2 + 48 * y - 36 * x * y + 27 * y ** 2))
return z
if __name__ == '__main__':
x = Variable(np.array(1.0))
y = Variable(np.array(1.0))
z = goldstein(x, y)
z.backward()
print(x.grad, y.grad)
def test_neg_overload(self):
x = Variable(np.array(2.0))
assert (-x).data == -2.0
def test_neg_backward(self):
x = Variable(np.array(2.0))
y = -x
y.backward()
assert x.grad == -1
def test_calc_multi_type_ndarray_right(self):
x = Variable(np.array(2.0))
y = np.array(3.0) + x
assert y.data == Variable(np.array(5.0)).data
def test_name(self):
x = Variable(np.array(1), "test")
assert x.name == "test"
class TestVariable:
def test_name(self):
x = Variable(np.array(1), "test")
assert x.name == "test"
@pytest.mark.parametrize("input, expected", [(Variable(np.array(1)), 'int32'), (Variable(np.array([1.0, 2.0, 3.0])), 'float64'), (Variable(np.array([[1.0, 2.0], [2.0, 3.0]])), 'float64')])
def test_dtype(self, input, expected):
assert input.dtype == expected
@pytest.mark.parametrize("input, expected", [(Variable(np.array(1.0)), 0), (Variable(np.array([1.0, 2.0, 3.0])), 1), (Variable(np.array([[1.0, 2.0], [2.0, 3.0]])), 2)])
def test_ndim(self, input, expected):
assert input.ndim == expected
@pytest.mark.parametrize("input, expected", [(Variable(np.array(1.0)), 1), (Variable(np.array([1.0, 2.0, 3.0])), 3), (Variable(np.array([[1.0, 2.0], [2.0, 3.0]])), 4)])
def test_size(self, input, expected):
assert input.size == expected
@pytest.mark.parametrize("input, expected", [(Variable(np.array(1.0)), 1), (Variable(np.array([1.0, 2.0, 3.0])), 3), (Variable(np.array([[1.0, 2.0], [2.0, 3.0]])), 2)])
def test_len(self, input, expected):
assert len(input) == expected
@pytest.mark.parametrize("input, expected", [(Variable(np.array(1.0)), ()), (Variable(np.array([1.0, 2.0, 3.0])), (3,)), (Variable(np.array([[1.0, 2.0], [2.0, 3.0]])), (2, 2))])
def test_shape(self, input, expected):
assert input.shape == expected
@pytest.mark.parametrize("input, expected", [(Variable(np.array(1.0)), "variable(1.0)\n"), (Variable(np.array([1.0, 2.0, 3.0])), "variable([1. 2. 3.])\n"), (Variable(np.array([[1.0, 2.0], [2.0, 3.0]])), "variable([[1. 2.]\n [2. 3.]])\n")])
def test_shape(self, input, expected, capsys):
print(input)
captured = capsys.readouterr()
assert captured.out == expected
@pytest.mark.parametrize("input, expected", [(np.array(1.0), Variable(np.array(1.0))), (Variable(np.array(1.0)), Variable(np.array(1.0)))])
def test_as_variable(self, input, expected):
assert type(as_variable(input)) == type(expected)
def test_pow_backward_overload(self):
x = Variable(np.array(2.0))
y = x ** 3
y.backward()
assert x.grad == 12
def test_pow_overload(self):
x = Variable(np.array(2.0))
assert (x ** 3).data == 8
def test_pow_backward(self):
x = Variable(np.array(2.0))
y = pow(x, 3)
y.backward()
assert x.grad == 12
def test_pow(self):
x = Variable(np.array(2.0))
assert pow(x, 3).data == 8
def culc_object():
yield Variable(np.array(2)), Variable(np.array(3))
def test_backward(self):
x = Variable(np.array(3.0))
y = fn.square(x)
y.backward()
expected = np.array(6.0)
self.assertEqual(x.grad, expected)
def test_neg(self):
x = Variable(np.array(2.0))
assert neg(x).data == -2.0
import numpy as np
# core_simple を明示的にインポート
# （dezero/__init__.py の is_simple_core = False でも動作させるため）
from dezero.core_simple import Variable
from dezero.core_simple import setup_variable
setup_variable()
def rosenbrock(x0, x1):
y = 100 * (x1 - x0**2)**2 + (x0 - 1)**2
return y
logs = []
x0 = Variable(np.array(0.0))
x1 = Variable(np.array(2.0))
iters = 1000
lr = 0.001
for i in range(iters):
print(x0, x1)
y = rosenbrock(x0, x1)
x0.cleargrad()
x1.cleargrad()
y.backward()
logs.append([float(x0.data), float(x1.data), float(y.data)])
def test_calc_multi_type_left(self):
x = Variable(np.array(2.0))
y = x + 3
assert y.data == Variable(np.array(5.0)).data