Request for Comment on MMR Refactor

[SpookedOne]

As some of you may be aware, the Trench Wars development team is working on some refactors to the current mmr system. There are quite a few changes so please review and leave any comments or concerns you may have.

Ratings

• Player ratings will be reset.
• Players start at 1500 instead of 3000.
• Estimated scores (which is used in end of round rating adjustments with actual score) uses a new algorithm that is more accurate.
• Wins and losses now factor in final score (before would assign 0.0 for loss and 1.0 for win)
  • Actual score is calculated by looking at total team deaths (for team elim) or flagtime (for base)
  • For example, winning 49-50 in team elim will net a score of 0.51 instead of the previous 1.0.
  • Finer score calculation also means it's possible to lose points even if you win (if not by margins in estimated score)
• New K-Factor (uncertainty) algorithm that scales with games played to quickly settle new players into a rating.
• Max rating change (unaccounting for K-Factor) is set to 32.0

Teams

• New greedy algorithm for building teams mainly based on highest rated players available (before used a brute-force shuffle).
• Optional flag for prioritizing bencher as well.
  • League ops will coordinate times when bencher prioritization will be disabled to get higher quality games.
• Players are sorted in following order, moving to next step if both players have the same value:
  • Benched last round (optional)
  • Rating (high-to-low)
  • Games played (high-to-low)
  • Date signed-up to TWD (earliest-to-latest)
• Teams are built by first team picking from sorted list of players, then compares each teams cumulative rating, and lower rated team picks next (possible for multiple times in row)
• Larger teams are prioritized over multiple smaller teams (ie, one 5v5 with 6 benched instead of two 4v4s if 16 queued)

MMRB

• MMRB is now supported as well.
• Players may only `!p` for one given ship at a time, so choose wisely.
• Ship are limitted to terrier, shark and spider for now.

Subbing

• New `!needsub` and `!sub` feature for requesting a sub or subbing in for another player.
• Subs must be within 200 rating point diff from person they are subbing.
• Estimated and actual scores for rating change is adjusted for each players time played.
• You may not use `!needsub` if you used `!sub` to get into a game.

Performance Blocks

• With the more accurate estimated and actual scoring for rating adjustments, performance blocks have been removed for the time being.
• Players can still be banned however for afking, teamkilling, abusing `!needsub` or using any other feature of the bot as not attended.

Automation

• Bot will be fully automated.
• Referrees will lock bots to given arena and a queue will start when a player uses `!p`.
• Queue will stay open for up to 5 minutes while players join.
• Timer for 30s will be set to make teams if queue reaches minimum number of required players (ie, 8 for 4v4)
• A shorter timer for 10s will be set to make teams if queue reaches maximum number of required players (ie, 10+ for 5v5)
• Teams will generate and be assigned to an arena. After last arena finishes it's round, the bot will reopen the queue after 10s.

[RaCka]

HYPE!! Nice work, Spooked!

[Rab]

I don't see how basing can work if it's one ship only, but the rest of it sounds excellent and long overdue. I think the benched priority flag won't be used.

[SpookedOne]

Only can queue for a single ship in base right now. You could play a shark one round then spid another. Multiship queuing will be a feature in near future but complicates calculating team sizes (ie, we need to know upfront you'll be selected to spid instead of shark based on all other players and only tally you for available spids)

[RageRitual]

A few early thoughts:


For base a "fill" option would be nice. A lot of us can play all 3 ships these days and having the option to fill into any ship that might be lacking could help get games going.

The sub feature would be nice if it announced in arena text if you need a sub and anyone within the rating can do command to sub in, instead of needing a host to do it.

Can we have set so players avoid being benched twice in a row? Considering literally every other arena is dead while MMR is going, everyone should get a chance to play. (unless blocked)

Since TWD has been replaced by MMR, is there any chance to just have MMR stats post to TWD website so we have a stats area?


Thanks for taking on this task regardless of how it works out.

[SpookedOne]

Yep, sorry if that was unclear but any player can use !needsub and bot will announce in arena for spectators to use !sub to jump in (within 200 rating allowance). Also, the team generator will prioritize benchers by default but we've allowed it to be disabled to allow for windows of "higher quality" play.

As for new MMR page/api, that is something on the roadmap but can't promise a date just yet. We have a database migration that will likely happen first then new API can be dropped in, and then likely a website to follow.

[RageRitual]

Yep, sorry if that was unclear but any player can use !needsub and bot will announce in arena for spectators to use !sub to jump in (within 200 rating allowance). Also, the team generator will prioritize benchers by default but we've allowed it to be disabled to allow for windows of "higher quality" play.

As for new MMR page/api, that is something on the roadmap but can't promise a date just yet. We have a database migration that will likely happen first then new API can be dropped in, and then likely a website to follow.

Awesome spooked!

If someone is eventually working on the website I have some ideas if whoever that ends up being is interested.

[Kov]

Good work on the new bot auto-starting after the queue reaches a certain amount to keep things going faster, love it.

Not sure about the ratings change, especially getting a negative rating change even with a win. Just had a 10-6 mvp win (weak mvp I admit yes), but still ended up with the most points lost in the rd. Don't think in any situation a 10 kill MVP should lose points, even if they were expected to get more based on the lineups. If that were the case, then the rating change seems to be too heavily weighted on the randomly generated lineup, and not weight enough toward in-game performance.

[SpookedOne]

I understand that having an mmr rating based on team performance (especially with score expectations) can be offputting, but given this is teamelim it's the most straightforward way to rate.