After great support and feedback from you guys, I've made some more changes to the algorithm and so REBBL Elo tries to predict your results once more!
Not all games have been played at the time of writing and the predictions results have been subpar. After using season 1 as input, trying to predict every subsequent game usually shows results around 50% of overall games and 63% of all non-draws. Results from the past week have been 43% overall and 56% of non-draws (of course, removing admin games from the equation), which are, respectively, 5% and 6% better than random picking (23% of games have been draws), but still around 7% worse than what we're looking for at the moment.
Not good enough.
For this week, though, the priority was to remove admin games from the collected data so as to not pollute findings. We want to predict real games, after all. Removing those, however, showed that the former chosen set of parameters was not the best performing and those have been substituted by the best I could find (still haven't implemented a scientific way of choosing parameters, other than good ol' Charles picking whatever and comparing results; this is were exciting genetic algorithms or machine learning would be an easy fit). That will explain why some rankings have changed quite a bit, as wins against admin teams have been removed from records.
Unplayed games from week 2 are not affecting rankings, obviously, which can cause some more prediction errors. We'll have to live with those.
Race matchup results have also been changed as humans dont get all of those admin losses and, as a bonus, GMAN games have been added. I also have predictions for all GMAN games now, but the process of doing the "pretty" tables is still manual so it would be too much work for me to do those.
I've encountered an issue while bringing BIG-O games over, because their season 1 corresponds to our season 4, it seems. So I've rolled it back to work on it some other time.
Please, keep the feedback coming and do reach out if you want to help the project (shout out to my boys DavidLeClair and tself55 for their input on the algorithm). Without further ado, these are the predictions for next week: