Smogon Forums

Hm, looks interesting. What sort of heuristics are you using for the suggestions?

It's based on Antar's checks and counters data.

It finds counters to your team and scores them based on how common they are and how well they counter your pokemon. Then it finds counters to those and scores them based on how well they counter the counters, and the original scores. Suggestions are ranked by that score.

If it was highly used, the weight by commonness would pick up things it suggested in the past, so it shouldn't stagnate. And if a new set comes up that counters an old one very well, it will rapidly make its way into the suggestions.

Interesting, and not a bad approach. It does mean that we'll need to embed up-to-date check/counter data.

That's true. I'd like to include the data generation as part of the scripts that update OU statistics. But there's quite some time between now and when the current data gets stale.

For more limitations: the current XY OU data file is 443k, and could be improved by making it bigger. I can save 33% of that for a minor speed hit. That's just one tier; each tier needs its own data file.

There's currently no way to generate the data for doubles formats. The data is currently useless in Ubers. The entire approach is invalid in less popular tiers.

Not much can be done about the size problem: if that's too big then it won't work. Doubles and Ubers could eventually be fixed. There's nothing to do about less popular tiers other than just turning it off.