Until Nintendo adds a 96 race grand prix, we are kind of stuck with this 2 x 48 analysis format, because as you may have noticed, drivers between the first and second 48 change. I completely understand why they haven't added this 96 race format, as this whole premise is kind of absurd.

Data Retrieving Methodology

This year, I wanted to see just what kind of data I could somewhat easily record and analyse from the 96er.

So, I thought that it would be a good start to take a photo of the scoreboard after every race (except the 2 I forgot, Cheep Cheep beach and Daisy Cruiser). From here, I was able to over engineer a solution to get the data from the photos into a format that I could analyse.

First, I did have to go through and manually name each of the 94 images I took with something resembling their track name. This took me about half an hour and turns out it wasn't really necessary, but it makes the input data a bit nicer so thats something.

Now that I had all of my 'TRACK_NAME.HEIC' image files:

It's time to send it to GPT to ask it to rip out the information I need. Unfortunately, as I was about to learn, GPT does not allow for processing of HEIC images:

so I had to write another script to convert the images from .HEIC to .PNGs.

which seems to have worked well:

Now that I can send the images to GPT to analyse, I had to cook up a prompt to get GPT to return the data in the format I wanted. I went for:

You are mario kart score GPT. Here is a photo of a mario kart scoreboard.

Can you please return this leaderboard as a JSON array of 12 items,
containing the player and score?

Please add a START and END string at the beginning and end of the JSON,
and do not add any formatting (new lines, tabs, etc).

I just needed to run this for all 94 of my newly converted PNGs and I should have all scoring data I need output to a single folder, and voila, it appears to run well:

Now that I have a folder full of the scores from each map, I can really cook.

First, I need to write a script that takes all of the JSON files, and extracts the greatest score value. This will help me sort the input data into a somewhat chronological order, which will be easier to plot along an x-axis.

But, here is where I hit another snag - as we did 2 x 48 races, I had 2 similar very score ranges. e.g, Mario Kart Stadium and Paris Promenade were the same values for max points score - and at the moment I didn't have a way to differentiate between which race was in the first 48 and which was the 2nd.

However, this is where I was able to repurpose the array I used in the table at the top of this screen. It's in the form:

const results = [
  ...
  {
    track: "Big Blue",
    winner: "Ben",
    insights: "Last Race wins (first half)",
  },
  ...
]

So, if I just rip out the track name and remove the spaces - I should be able to match most of the track names to the score data file I have for that track. This initially worked pretty well, basically leaving only duplicate rainbow roads and the 2 maps I forgot to take photos of:

From here, I just had to add an index the rainbow road results array to match how I stored the photos of the Rainbow road scores:

here,

• RainbowRoad_Blue.png - the rainbow road with the tunnel where you should hit the blue at the end
• RainbowRoad_hardturn.png - the rainbow road with the super hard turn at the start
• RainbowRoad_Jumps.png - the simpler rainbow road with the little jumps
• RainbowRoad_NoLaps.png - the rainbow road where it's all continuous, no laps
• RainbowRoad.png - the first file I named so it just got called rainbow road

Once I factored this in, I was able to sort the race data I had from races 1-96 (except the 2 I forgot):

Once I had this, I just had to write this script that outputs the track placing data (our .jsons) into a format that I could use to plot the data on a graph, e.g:

const results = [
  {
  track: 1,
  "Donkey Kong": 15,
  "Black Shy Guy": 12,
  Bowser: 10,
  Toadette: 9,
  "Baby Daisy": 8,
  Waluigi: 7,
  Luigi: 6,
  "Pink Gold Peach": 5,
  "Tanooki Mario": 4,
  "King Boo": 3,
  Link: 2,
  Kamek: 1,
},
{
  track: 2,
  "Donkey Kong": 30,
  "Black Shy Guy": 21,
  Waluigi: 19,
  Toadette: 19,
  Bowser: 17,
  "Pink Gold Peach": 13,
  "Baby Daisy": 13,
  Luigi: 12,
  "King Boo": 7,
  Link: 5,
  "Tanooki Mario": 5,
  Kamek: 3,
},
...
]

And, that's how I got the above line graphs. I only budgeted 4 hours to this whole analysis and am already sick of looking at mario related data, so I'll leave it at that.

See you next time.