zkstats: 1v1 Matchmaking Statistics

sort
esainane 5 years ago This has been running for a while now, but I've only recently been reminded that I neglected to also post this on the forums. https://zkstats.antihype.space/ The source is available at https://github.com/esainane/zkstats/ Almost all elements are clickable and will filter other dimensions. Map categories were added very recently and are quite experimental. The mappings can be found at https://github.com/esainane/zkstats/blob/master/public/data/map-types.json - any suggestions or corrections would be more than welcome. Suggestions or corrections in general would also be more than welcome, of course. +17 / -0
DeinFreund 5 years ago (edited 5 years ago) Funny how you can see the second factory transition from Strider Hub to Gunship to Air as Elo increases. Really nice interface btw! +3 / -0
esainane 4 years ago There was an unusual replay blocking updates. This has now been fixed. Regular processing should finish catching up on the backlog and resume in about four hours. +2 / -0
nop 4 years ago This is really cool, it's nice to be able to point to a graph to show how balance interactions are working or not working. Would be nice to see charts of metal spent on each unitdef in winning/losing games, to identify which units are monospammed or are not being used. +0 / -0
Shadowfury333 4 years ago While I'm not much of a statistician, I figured I'd take a bit of a look. Some random findings looking at the last 3 months: Looking at 2200+ elo with a +/-200 elo advantage for the winner (still kinda even skill) Just about every factory is viable somewhere, with a slight advantage to Shieldbots vs Any (55-45), but overall everything is fairly even. Spiders was surprisingly viable against Rovers, though. Looking at ~1600 elo (the biggest elo block on the histogram) with a +/- 200 elo advantage for the winner, Rovers have a 75% chance to win, while Shieldbots, Amphbots, and Tanks also have an advantage, while Hovercraft doesn't work (probably because Daggers take a lot of practice to use well). At the highest levels Shieldbots become quite reliably useful, but everything is overall still pretty close to 5-5 matchups. The first 3 months of stat collection also show much stronger matchup advantages along all of the above lines, so I guess the matchup stats present a positive trend in game balance. +4 / -0
esainane 4 years ago Implemented and added experimental scaling of each row and column in the matchups chart by relative popularity. You can now: Watch the popularity of Rover explode in mid-high ranks, then quickly drop off above that. Watch Hover instantly get discarded at all levels as soon as you look at maps with even light hills. Boggle at Cloaky actually appearing to become slightly less popular, relatively speaking, since the big patch hit. ...really? The box within each row/column is still scaled to indicate lack of significance if there are very few matchups available, which is possibly redundant with this new presentation. The Matchups chart now conveys an extremely dense amount of information. If you want the old behaviour, type the following into the console: m=charts.get('matchups') m.sizeByPopularity(false) m.redrawGroup() +1 / -0
esainane 4 years ago Also, FrostyCove v1.11 and Vantage v1.0 are now classified as Light Hills, rather than Hills. Vantage v1.1 has been added, classified as Light Hills. +1 / -0
esainane 4 years ago (edited 4 years ago) Blockage on B836420 2 on Gecko Isle 1.1 fixed. +1 / -0
esainane 4 years ago Added the new matchmaking maps. Someone yell at me about adding a new "Cliffs" map type category if that's not done in a day or two. +1 / -0
Thomas1 4 years ago yells in conquistador +0 / -0
esainane 4 years ago We've now hit over 10k matches processed! [Spoiler] Pay no attention to the hundreds of "Unknown" matches behind the curtain. That'll be fixed soon >_> +3 / -0
esainane 4 years ago The recent blocking issue should be fixed. People can also now check a status endpoint to see if things are currently broken. Ignore the off-by-one for now. There's also a done-when-its-done page for tracking the most basic facets of team games, which is currently showing the relative popularity of group balance teams versus solo balance teams throughout the day. Not terribly usable, no labels or anything remotely friendly, but it should still be an improvement over manually polling the replays page. +6 / -0
malric 4 years ago (edited 4 years ago) Could you please mention here at least what is blue (top) and what is yellow (bottom) for tracking the most basic facets of team games? +0 / -0
Aquanim 4 years ago Group balance is in blue on the top I assume since there is no yellow earlier than a week or so ago. +1 / -0
Berder 4 years ago Fascinating! +0 / -0
Berder 4 years ago It's interesting that if you restrict the loser's elo rating to 2200+, the first factory with the highest performance is the Gunship Factory. It seems to be a pattern of using a lot of Locusts to out-raid the opponent on maps where the opponent might use hovercrafts: Adansonia, Leyline, Firebreak, Crubick Plains. Not necessarily even involving an early fac switch. +0 / -0

Forum index > General discussion >