WIP: Add layout database

[jbdyn]

This PR adds a Python script for generating the database for Pelita mazes.

[Debilski]

I refined the existing walls_to_graph implementation in #853 so that it can be reused here.

[jbdyn]

I refined the existing walls_to_graph implementation in #853 so that it can be reused here.

Oh, sorry, I didn't see that. I also reworked the code in create_db.walls_to_graph.

[otizonaizit]

I refined the existing walls_to_graph implementation in #853 so that it can be reused here.

Oh, sorry, I didn't see that. I also reworked the code in create_db.walls_to_graph.

do you have a profiling for this? How much faster is it to start with a fully connected graph and remove the nodes?

[Debilski]

I just profiled it. @jbdyn’s new variant is twice as fast as ours on the normal mazes. Interestingly, nx.grid_2d_graph does not apply any magic at all (if fills the whole thing with Python for loops), so this is probably a case of simply choosing the right iterator for the right data structure. (see: https://networkx.org/documentation/stable/_modules/networkx/generators/lattice.html#grid_2d_graph )

[Debilski]

def find_chambers(G: nx.Graph, shape):
    w, h = shape

    main_chamber = set()
    chamber_tiles = set()

    for chamber in nx.biconnected_components(G):
        max_x = max(chamber, key=lambda n: n[0])[0]
        min_x = min(chamber, key=lambda n: n[0])[0]
        if min_x < w // 2 <= max_x:
            # only the main chamber covers both sides
            # our own mazes should only have one central chamber
            # but other configurations could have more than one
            main_chamber.update(chamber)
            continue
        else:
            chamber_tiles.update(set(chamber))

    # remove shared articulation points with the main chamber
    chamber_tiles -= main_chamber

    # combine connected subgraphs
    subgraphs = G.subgraph(chamber_tiles)
    chambers = list(nx.connected_components(subgraphs))

    return chambers, chamber_tiles

Using this instead of paint_chambers reduces the time from 9.3s to 6.5s on my system. (We can now also skip searching for the cuts before.)

[otizonaizit]

@jbdyn : is it OK if we push to this branch to move it forward?

[jbdyn]

is it OK if we push to this branch to move it forward?

Sure!