def test_sliding2d():
"""
Input: A (padded) valid Tensor for slides. [width, height]
width_idx
height_idx
stride
Output: A sliding Tensor [K, K], K = kernel_size
"""
n_samples = 2
width = 4
height = 5
a = Tensor(np.random.randn(n_samples, width, height), requires_grad=True)
a.print()
kernel_size = 2 # Symmetric Squared
stride = 1
width_idx = 0
height_idx = 1
b = slide.Sliding2D(width_idx=width_idx,
height_idx=height_idx,
kernel_size=kernel_size,
stride=stride)(a)
b.print()
b.backward()
print(a.grad)
def test_padding2d():
n_samples = 4
width = 2
height = 3
a = Tensor(np.random.randn(n_samples, width, height), requires_grad=True)
a.print()
padding = 1
b = pad.Padding2D(padding=padding)(a)
b.print()
b.backward()
print(a.grad)
def test_mse_loss():
n_samples = 5
n_output = 4
Y = Tensor(np.random.randn(n_samples, n_output), name='Y')
Y_pred = Tensor(np.random.randn(n_samples, n_output),
requires_grad=True,
name='Y_pred')
loss_ = mse.MSELoss('loss')(Y, Y_pred)
Y.print()
Y_pred.print()
loss_.print()
loss_.backward()
print(Y_pred.grad)