Embedded nodes in draw_multilayer

[thomasrobiglio]

Adresses #436. I am still working on this to try to change as little as possible to behavior of the function.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[maximelucas]

Quoting @thomasrobiglio :

Yes I thought the same about the nodes' shape. What can be more relevant is the following:

I am struggling to have a node collection returned when calling the function (which then allows to have the colorbars and the cool stuff we have for the standard drawing function).

This is because I need to create a separate Circle patch that is transformed into a 3D patch for each node (in each level) separately, and I have not managed to group these objects in a collection so far. I will keep looking for a solution, if you have ideas they are most welcome, I am a bit stuck--I have opened a draft pr #612

Yes, I thought that might be a problem. Matplotlib has different types of collections, including a PatchCollection with an example here. Would this work with the 3D patches?

[thomasrobiglio]

Thank you @maximelucas! this is more or less the documentation I was looking at when working on this... let's see if I can find something that works :).

[thomasrobiglio]

This could be a way out... to have the circles for the nodes drawn as polygons with many sides. It's not the best thing to do, but I could not find a way to use a collection of circle objects in 3d.

With a couple of tweaks, I should be able to make it work from outside as the function we have in the stable version now.

[maximelucas]

Could you show us what the viz would look like? Is it like your example in the issue you opened?

[thomasrobiglio]

Something like this---if you have more >20 points to build the polygon you cannot see that it's not an actual circle:

[maximelucas]

@thomasrobiglio nice, even if we view it from above? Do you think the colors of the nodes are still visible enough (for example when we plot stats)?

[thomasrobiglio]

thanks for the ping @maximelucas! tbh this had fallen into oblivion on my to-do list... I'll try to make some further progress

[maximelucas]

Apparently you could do something like this instead:
Build real Circle patches (true circles, no polygon hack), put them in one PatchCollection with the colormap-array, then convert to 3D (with art3d.patch_collection_2d_to_3d), mirroring the pattern of edge_collection already uses at draw.py:843:

  # one Circle per node per layer
  patches = [Circle((x, y), radius) for ... ]
  node_collection = PatchCollection(
      patches,
      array=node_fc_arr,      # values → mapped by the collection (enables colorbar)
      cmap=node_fc_cmap,
      edgecolors=node_ec,
      linewidths=node_lw,
  )
  art3d.patch_collection_2d_to_3d(node_collection, zs=z_per_patch, zdir="z",
                                  depthshade=False)   # keep stat colors faithful
  ax.add_collection3d(node_collection)

Pros compared to current PR solution:

1. Real circles, not 30-gons.
2. One ScalarMappable collection → plt.colorbar(node_collection) and stat-based coloring work, like the 2D draw_nodes.
3. You can collect circles across all layers into a single collection (pass per-patch zs), so draw_multilayer can
   return one node_collection matching the 2D API, instead of an unreturned per-layer collection.
4. It reuses the exact idiom already in draw.py (the edge_collection block), so it's consistent.
5. depthshade=False keeps colormap colors accurate.

Worth having a look