Updated most recent period to include matches up to 22nd July 2022
Breaking - Found and fixed a critical bug that meant that the “averaged win rates”, “civilisation v civilisation win rates” and “hierarchical clustering” outputs were highly inaccurate. I apologies for any inconvenience caused to anyone who may have previously used or relied upon these outputs
Added new filters for Nomad (both solo and team) to the most recent period
Updated the highest Elo filter to only include matches for >1700 Elo (was previously >1600)
Added a new “all maps” filter for the >1700 Elo group
Updated filters for team games to use the new Elo scores (centred on 1000)
Retrofitted the new Elo calculation to historic games. Removed team games that occurred between the 14th and 15th July due to inaccuracies in the data of when the conversion was applied
Reduced the arbitrary reference lines on team win rate plots from +/-5% to +/-3% to better represent expected normal ranges for team games
Added new output “Median Number of Matches Played per Player by Elo Percentile” to the Global Statistics page
Updated the output title “Distribution of Matches Played by Mean Team Elo” to “Distribution of Matches Played by Mean Match Elo” to better represent the output
Changed methodology for how averaged win rates and their confidence intervals are calculated; mostly impacts team games, see methodology tab for details
Increased minimum time cut-off for closed maps from 6 minutes to 9 minutes
Thanks for your work, but I want you to know that there are two instances of “RM, Solo, Open > 1700 ELO”. The first is obviously the new “all maps” (which I really like as addition), but you might want to correct the name to avoid confusion.
Hmm, now that I closely see them, it appears that the two are duplicates, check this. Left is all maps, right is open. They seem identical, is it normal?
Comments on the data: Bengalis are truly terrible, #2 lowest win rate even on team closed maps, where they are supposed to shine. Dravidians at least dominate Nomad team games and I can see why.
Thanks for highlighting this, this is just an error from me being tired and copying the config block and only remembering to change the id and not the actual parameters. Will update and push the fix this evening. Thanks for spotting !
Just to say I applied the fix to the website. I also spotted a bug in the sort order for the individual civilisation v civilisation plots which I fixed. There is a chance that you may need to reset your cache to view the new plots. Please let me know if you spot anything else !
Yeah them and port are a difficult one to balance if we want them more appealing on Arabia without making them broken where they’re already good.
In port’s case is it literally just down to the organ gun? That it’s so good on a map like arena. But not good enough that the civ can revolve around it on Arabia or even nomad?
Organ Gun strat is harder to deal in team games (just as the old arambai and pre nerf war wagon) than 1v1, and is right now (after the projectile speed buff) is increasing the civ pickrate and then WR for closed maps TGs.
Best thing to do is just remove one range, so they can’t just sit and destroy a TC safely, in compensation then the elite organ gun gives +2 range and +1 pierce armor, so they still are worth to use in Imperial.
I don’t know how to buff Malay, but what Ports need is a early-game buffs certainly. This would keep their goodness on water maps because the bonus would’t be maritime, and helping their weakness on the first period of the game.
But that’s the core of the problem, Portuguese late game is quite strong and is balanced by a weaker early game, you can’t have a civ strong in all ages.
But at this point the civ has 4 bonuses, more is overkill.
Also, I feel Portuguese is just not explored enough, they can:
Do a 4 militia drush.
Have a nice tech advantage when going for archers thanks by the 30% faster researched techs.
Having cheaper gold units means you don’t need many villagers on gold.
30% fast researched wheelbarrow gives you one villager advantage.
In castle age they can go for more than one gold unit quite easily thanks by both gold discount and fast researched techs.
In Imperial you can change the army comp nearly as good as chinese, also, you get chemistry faster.
And lastly, you just can’t overcharge a civ with too many bonuses, Tatars are a good example of this, back in 2020 the civ was deemed way too bad and too weak, in the whole year got a lot of buffs, until in the 1st anniversary patch they got the free sheeps in feudal age and new TCs in castle age + better silk armor, what happened? Tatars became overpowered on open maps (I mean, Fatboy was climbing the ladder so hard with just Tatars), till LOTW release when the civ got nerfs (no more free sheeps in feudal and keshik more expensive), since then, most pros started to use them in tournaments and overall aren’t seen as a bad civ anymore.
Apologies just realised I broke the triggers for updating the cross plots so they currently only update when you click to open the dropdown not when you actually select something. Apologies will try and fix this tomorrow evening (will also try and start looking more into JS automated testing frameworks ).
I had a couple of requests for this as it’s a different way to consider time. If anything I would personally argue it’s slightly more intuitive than the within variant as in it can be thought of as “given Ive made it to minute x what’s my probability of winning”
Yer this is reasonable / sensible. Will add it to the to-do list
I’ve been thinking about this a lot and can’t make up my mind. Originally it was just >1200 because that’s what I felt best represented the state of the game both from a civ performance and from a what people are really experiencing point of views. Then by demand I added in the >1700 filter for people just wanting higher Elos. Then due to even more demand I added 900-1100 filter but now the >1200 looks a bit odd given it overlaps with the >1700 but then I don’t want to get rid of it as it’s original intent still stands. I also don’t want to add in 1200-1700 as well as the existing one because the computation processing is getting pretty intense already (currently takes about 3 hours to rerun stuff when I make a change). So yer going in circles on this one…
I’m probs going to leave this in as I got a lot of requests for it from people who wanted to see the win rates at their Elo (as this filter covers something like 40% of the playerbase).
As an aside I silently released an update though didn’t really promote it because the changes weren’t that major. Main thing was a complete rework of the backend data base which makes it much easier for to backup and transfer the data. O I also relaxed my data inclusion rules to allow for maps with an unknown version and also added some minor imputation of players with missing Elos. These 2 changes lead to a 10% increase in sample size.
Also of note the updates are likely to slow down for quite a while sorry as we are about to have our first child. I hope to keep up with refreshing the data but finding time for substantial changes is probs going to be hard for a while.
).