I'm not sure if this is the intended behavior or not. nn.Module does not look for parameters inside lists. This makes it hard to e.g. initialize a network, provide the number of layers as an argument, and then store these layers in a list.

For example:

class net( nn.Module):
    def __init__( self, num_layers):
        super( net, self).__init__()

        self.layer = nn.Linear( 1, 1)
        self.list_of_layers = [nn.Linear( 1, 1) for i in range( num_layers)]

n = net( 4)
print n

Returns:

net (
  (layer): Linear (1 -> 1)
)

I would have expected the parameters in self.list_of_layers to be have been inherited as the network parameters too.

Is there another recommended way that allows storage of an arbitrary number layers?

Thanks