Adding nn.Narrow with pytorch

How can I do the same thing with pytorch?

local right = nn.Sequential()
  right:add(nn.Narrow(2, 1, output_size))

You could write a custom module and pass add it to nn.Sequential.
Something like this might work:

class Narrow(nn.Module):
    def __init__(self, dim, start, length):
        super(Narrow, self).__init__()
        self.dim = dim
        self.start = start
        self.length = length
    def forward(self, x):
        x = torch.narrow(x, self.dim, self.start, self.length)
        return x
1 Like

Are there any ways to hold this tensor input empty?

I don’t quite understand the use case. Could you explain it a bit more?

So I could later create concattable and assign same tensor input to this module and others.

I still don’t understand the use case. You are not assigning a tensor input to the module, but create the module and pass the input during the forward pass to it.