What are the equations it uses model(x) manually?

I have extracted bias and weight from trained model.

print('X = ',X[0,:])
print(type(pinv(Wy)),pinv(Wy).shape)
Wy.detach().numpy()
h1 = torch.matmul(Wy.T,y2_pred-by)
k1 = torch.tensor([math.tanh(h1[i]) for i in range(len(h1))] ) ; 
x1 = torch.matmul(Wx.T,k1-bx) ; print('X1 = ',x1)

but give this result:

X =  tensor([-0.7951, -0.0736,  0.8450,  0.8444,  0.8469, -0.9261, -0.9261, 12.0541])
X1 =  tensor([ 0.4630, -0.3566,  0.8700,  1.3055,  0.0930, -0.2514, -0.1384,  0.0268]

Where is the problem and what are all equations for reproduce the model(x) command.??
Im using:

  • One hidden layer
  • activation function = tanh
  • x=3 h= 6 y = 8

Your code tries to use atanh instead of tanh no?

i change with correction

Do you have the code that defines the model? In particular the __init__ and forward functions.