tar_y = Tensor(v)
mse = MSE_Loss()
for e in range(0):
y = conv2d.forward(input,
k,
kx=kx,
ky=ky,
kz=kz,
channel=channel,
stride=(sx, sy, sz))
loss = mse(y, tar_y, (0, 1, 2, 3))
loss.backward()
k.value -= k.gradient * 0.000001
#print(k.gradient)
k.zero_grad()
#input.value -= input.gradient * 0.001
#print(input.gradient)
#input.zero_grad()
print(e, loss.item())
print(tar_y - y)
print(input)
print(k)
from np.ad.operation import conv2d_transpose
cx = 2
cy = 2
cz = 2
[24.0, 32.0],
[48.0, 56.0]])
tar.append([[24.0, 32.0],
[48.0, 56.0],
[48.0, 64.0],
[56.0, 112.0]])
target = Tensor(np.array(tar))
mse = MSE_Loss()
for e in range(30000):
y = conv1d.forward(input, k, kx=2, ky=2, channel=2, stride=(1, 1))
loss = mse(y, target, (0, 1, 2))
loss.backward()
k.value -= k.gradient * 0.0001
#print(k.gradient)
k.zero_grad()
input.value -= input.gradient * 0.001
#print(input.gradient)
input.zero_grad()
print(loss.item())
print(y)
print(input)
print(k)
from np.ad.operation import conv1d_transpose
inp = []
inp.append([[1.0, 1.0, 1.0],