Better explanations re section names in run_rib_build and collect_interaction_edges

[stefan-apollo]

I am confused about this in run_rib_build.py

Don't build the graph for the section of the model before the first node layer

And this in collect_interaction_edges

Note that there is no edge weight that uses the position of the final interaction matrix as a starting node. This means that, unless node_layers contained the model output, we ignore the final section name in section_names when calculating the edges.

Just logging confusions to clear up later

[stefan-apollo]

Explanation I got from Dan, example cases to help understand why these two lines of code:

section_names = [f"sections.{sec}" for sec in seq_model.sections if sec != "pre"]

edge_modules = section_names if Cs[-1].node_layer_name == "output" else section_names[:-1]

1. node_layers = [mlp_out.3, output] --> expect graph spanning the part between mlp_out.3 and output --> model has sections pre{...}, sec0{...} --> we care about the sec0 module as the relevant function --> section_names = [sec0] --> edge_modules = [sec0]
2. node_layers = [mlp_out.3, mlp_out.5] --> expect graph spanning the part between mlp_out.3 and mlp_out.5 --> model has sections pre{...}, sec0{mlp_out.3...}, sec1{mlp_out.5...} --> we care about the sec1 module as the relevant function --> section_names = [sec0, sec1] --> edge_modules = [sec1]

[danbraunai-apollo]

I haven't really improved this documentation much. though I did add a single example in the rib_builder.py docstring (4de65d0). Hoping this is sufficient.