Hi everyone,
I’m working on an electromagnetism project using neural network and I’m trying to create a nn.Module class for a neural network for which many layers are contained in a dictionnary. Like this:
class ComplexNet(nn.Module):
def __init__(self):
super(ComplexNet, self).__init__()
self.R = dict()
for pp in ['00', '01', '10', '11']:
for ii in ['x','y','z']:
for jj in ['x','y','z']:
for kk in ['r', 'i']:
self.R[pp+'e'+jj+'e'+ii+'_'+kk] = HadamardProduct(Nky, Nkx, pp+'e'+jj+'e'+ii+'_'+kk, bias = False)
self.R[pp+'h'+jj+'e'+ii+'_'+kk] = HadamardProduct(Nky, Nkx, pp+'h'+jj+'e'+ii+'_'+kk, bias = False)
where HadamardProduct is a layer for dot product operation:
class HadamardProduct(nn.Module):
def __init__(self, Nky, Nkx, string, bias=True):
super().__init__()
self.Nky = Nky
self.Nkx = Nkx
self.weight = nn.Parameter(torch.load(f'{path}/R{string}')).double()
def forward(self, input):
return input * self.weight
The problem arises when I want to insert model.parameters() in the optimizer. It doesn’t recognize the different layers. Do you have any clue how to fix this issue?
Otherwise, I’ll need to create layers like this:
self.R00exexr = HadamardProduct(Nky, Nkx, ‘00exex_r’, bias = False)
self.R00exeyr = HadamardProduct(Nky, Nkx, ‘00exey_r’, bias = False)
self.R00exezr = HadamardProduct(Nky, Nkx, ‘00exez_r’, bias = False)
…
But there are 4x3x6 different ones.
Any help would be greatly appreciated!
Victor
