ssbu-gourmet-race-search uses the Bellman-Ford algorithm to calculate the optimal path in the Super Smash Bros. Ultimate Gourmet Race mini-game.

Here is the Gourmet Race map:

Here is the same map with the food visible. I couldn't find a higher resolution image and didn't want to build one myself by stitching together screenshots.

To view the algorithm running interactively, you can go to this Bellman-Ford graph algorithms website by the Mathematics Technical University of Munich. To use the interactive website, follow these steps:

1. Click "Test the algorithm!".
2. Upload the gourmet_race.txt file. You should see the following graph:

The edge weights correspond with the 1 point given per food, but they are negated because the Bellman-Ford algorithm finds the minimum cost path. You can determine the edge weights by comparing with the graph above.

Vertices are only placed in locations of the map that you have a choice. For example, between vertices n9 to p, the edge weight was given a cost of -3 even though there are intersections because the game does not stop and give you a choice of direction. There is only a single choice of direction because you can't walk backwards.

Here is the map labeled with the vertex names:

There are these perpendicular letter "H"-shaped paths that connect parallel roads. Like so:

They look like this in graph form:

The reason the cdef subgraph is made of 4 vertices while the hj subgraph is only made of 2 vertices is because of negative cycles. The Bellman-Ford algorithm can deal with negative edges but not negative cycles. If the cdef graph was like this instead, the Bellman-Ford algorithm wouldn't work because you could walk back and forth and get a negative infinity cost (infinity score).

So instead, the cdef subgraph is used so that you can only travel one of the -1 edges at most once. You can skip it by travelling the 0 edges. de and cf vertex pairs are basically the same location on the map but when travelling d->e or c->f, you are committing to skipping the H-path.

The hj subgraph has a cycle, but it is fine because it is not a negative cycle. There is no need to convert it to the cdef formation. If we did convert it, it would still work, but the algorithm would run slower because more vertices means more iterations for Bellman-Ford.

3. Note where vertex a is on the graph. a is the start and p is the end.
4. Click "Ready - Run the Algorithm!".
5. Choose vertex a as the start.
6. You can now click "next" to go step by step or "fast forward" to go faster. Note that it takes a while to run because it renders the steps for you to view.
7. After it finishes, you should see the following solution:

Notice how the optimal score is -11 for the finish vertex p. You can follow the green edges backwards from vertex p until you get to vertex a to see the optimal path.

The algorithm can run non-interactively and much faster on Mathematica.

The gourmet_race.wl file has the wolfram language code.

You can see the wolfram notebook. If you have a wolfram account, you could "Make Your Own Copy" to play around with the code interactively. If you do not have an account, you can do the following:

1. Go to the wolfram cloud website.
2. Click "Create a New Notebook".
3. Paste the contents of gourmet_race.wl into the notebook.
4. Press SHIFT ENTER to run the code.

You should see this initial graph:

and this solution graph:

The vertices are labeled 0-34 instead of a-z then n1-n9. But the number order matches the letter order. The labeled map above was labeled with both letters and numbers, so you can use it to match with the graph.

The optimal path given by Wolfram is {0, 1, 3, 5, 8, 16, 19, 26, 28, 21, 22, 20, 12, 13, 15} with a cost of -11.

The png, jpg, svg, and gimp xcf images are stored on google drive.