Draw trivial tokens and indices

[emileferreira]

Presently, drawing tokens or indices by themselves (for example, draw(tokens) or draw(indices, "hello")) results in an empty figure. This adds support for drawing trivial tokens and indices, as feedforward layers.

[gailweiss]

this draws them, but mysteriously labels them with FF in the drawing (instead of X, as they are labeled when drawing computation flows with actual computations). is it possible to get them to show up the same way they would normally?

[emileferreira]

They're now labelled 'X' in the computational flow figures. It's not an elegant solution, but the existing code only considers feedforward layers and would otherwise need to be rewritten.

[gailweiss]

if i understand correctly, you've treated them as their own parents, forcing them into the X row. it seems like a nice solution to me!

Can you add comments to the changes you've made, so it's clear why this special case (len(ffs)==1) exists now? i.e. specifically highlight around all your changes that len(ffs) == 1 implies we are drawing just a tokens/indices sop by itself, and (if it's confusing, eg switching ffs and their parents) explain what (and why) you are doing in this special case

[emileferreira]

Thanks. I've added comments to explain the changes.

[gailweiss]

it's a nice trick, but i don't like the use of the 'constant' patch as it permanently marks that sequence as a constant, which later creates strange behaviour (e.g. if you do draw(indices+1); after having earlier done draw(indices);, you will now get a warning that it is drawing a constant). however, the parent-child swap for the base sequence draw was quite nice. i have made a slightly different set of changes, based on your own, which do not cause this issue. i will note them in comments on your code

[emileferreira]

I figured that it was safe to (ab)use the _constant property of UnfinishedSequence, as it's unused besides for that benign warning. I've resolved the warning by resetting the property after drawing.

[emileferreira]

The trivial case, in which the feedforward layers and their parents are to be swapped, is now identified by the length of the respective lists. This avoids using the _constant property and triggering the associated warning.