import torch
import torch.nn.functional as F
def main():
# link rule
# x0 = torch.tensor(1)
# w1 = torch.tensor(1., requires_grad=True)
# b1 = torch.tensor(1.)
# y1 = x0*w1+b1
# w2 = torch.tensor(1., requires_grad=True)
# b2 = torch.tensor(1.)
# y2 = y1*w2+b2
# dy2_dy1 = torch.autograd.grad(y2, [y1], retain_graph=True)[0]
# dy1_dw1 = torch.autograd.grad(y1, [w1], retain_graph=True)[0]
# dy2_dw1 = torch.autograd.grad(y2, [w1], retain_graph=True)[0]
# print(torch.eq(dy2_dy1*dy1_dw1, dy2_dw1))
# single output perceptron
# X = torch.rand(1,10)
# W = torch.randn(1,10, requires_grad=True)
# O = torch.sigmoid([email protected]())
# print(O)
# t = torch.ones(1,1)
# E = 0.5 * F.mse_loss(O, t)
# E.backward()
# print(W.grad)
# print((O-t)*O*(1-O)*X)
# print(torch.eq(W.grad, ((O-t)*O*(1-O)*X).detach()))
# multi output perceptron
X1 = torch.rand(1,10)
W1 = torch.randn(5,10, requires_grad=True)
O1 = torch.sigmoid([email protected]()) # [1,5]
t1 = torch.ones(1,5)
E1 = 0.5 * F.mse_loss(O1, t1)
E1.backward()
temp = ((O1-t1)*O1*(1-O1)).detach() # [1,5]
print("X1:\n{}\nW1:\n{}\nO1:\n{}".format(X1,W1,O1))
print("W1.grad:\n{}".format(W1.grad))
print("W1.grad`:\n{}".format((X1.t()@temp).t()/5))
print(torch.eq(W1.grad, (X1.t()@temp).t()/5))
pass
if __name__ == '__main__':
main()
