special handling of #input == 1 about nngraph HOT 4 CLOSED

Because there are many nn modules that just expect a tensor. The default behaviour in nngraph is everything is a table of something. You can see here (https://github.com/torch/nngraph/blob/master/gmodule.lua#L171) that non-table inputs are put into a table. It made the logic in the in the rest of the system consistent to assume that every input is a table.

from nngraph.

An alternative consistent implementation could see everything as a nested table.
I.e., the input can be a tensor or a table with other tables or tensors.

from nngraph.

The inconsistency does make trying to understand gmodule.lua quite a lot more complicated than it might otherwise be. Lots of edge-cases and stuff. Would making the nn modules document what they expect as input help? Like:

function Linear:expectsTensor()
  return true
end

from nngraph.

I think we have resolved this offline already/

from nngraph.