Thinking back on the debate about what "good" is, I think 400+ QM rating is probably a pretty good litmus test.

Up till 400 you tend to gain as much as you lose per win/loss, so to hit 400 you need to win 50%. To rise past 400 though requires winning at least twice as much as you lose. I'd consider anyone that can win 67% without cherrypicking to be "good". Hitting 400 though is easy, so I'd pick some arbitrary number over 400 to ensure said people can actually win 67%, but even 500 or should be sufficient since to hit 500 you need to have won at least 67% over a span of 20+ games.

There are 18,018 accounts ranked on QM, only 222 have broke 500 so far.

508 people have hit 400, so more than half that have been able to win 50% over 60+ games have been unable rise from 400 to 500.

Edited 2/26/2018 18:10:17

I've hit 500+. I'm not good. Checkmate.

MDL rating is a much better litmus test. There's sufficient variation in templates that you can't do well just by optimizing your gameplay for the quirks of one template or a small collection of templates. There's limited options available for opponent selection or even template selection (basically just vetos); moreso than in any other competitive event, you need actual strategic and analytic ability (or intuition, depending on how you look at Warzone) to perform well on the MDL. Plus the MDL has a much more dominant presence of elite players (who, based on participation data, seem to enjoy the MDL more than they do the 1 v 1 ladder) so it's not like the 1v1 Ladder where a key factor in how well you do is which other players have decided not to participate.

The templates on the MDL are dissimilar enough that you can organize them (even with flawed data)
into coherent and relatively consistent similarity clusters (based on player performance between them):

The only issue with using the MDL here is that the participation is low, but in an ideal world I think we'd all be speaking about skill in terms of ratings on something analogous to the MDL (but not QuickMatch). I'd personally break down my tiers like this:

2000+ (95% expected win rate vs. average player on MDL): elite
1800+ (85%): great
1650+ (70%): good

Or you can look at eliteness as being so good that competitive events struggle to pair you with opponents of your own caliber. That cutoff is somewhere between 1950 and 2050.

Edited 2/26/2018 18:25:34

You're not great, but by most standards that aren't elitists posturing you're good.

If you are below 400 you win 10 points and lose 5 points against an opponent of your rating.

Edited my post with further discussion, but my point is that but for lack of participation, the MDL would be easily our best option for figuring out where players stand in terms of strategic ability.

I think we should use it as the end-all-be-all instead of the 1v1 Ladder like Wally is attempting to do.

Using a single-template ladder like the 1v1 Ladder is flawed due to the limitations of general Warzone ability. One thing I think we've mostly avoided here is that Warzone settings can get really diverse, even if you stick to what's generally considered strategic. And you can be good at one thing without being good at everything else; the fewer templates you use, the less confidently you can say that your results reflect actual general strategic ability (this is just intuitively true, although it's also supported by data).

On the Multi-Day Ladder, as players get better they also get (generally) less consistent across templates:


(note that r^2 is going to be small here even if there's a strong correlation purely because there's a lot of noise to deal with due to sample size issues- instead, it's better to just look at this visually and focus on the clearly visible linear growth pattern you see as you look left to right at most of the datapoints)

At the same time, though, it's not accurate to say that these great players are mostly doing well on a handful of templates and that's what's driving them up. While it's true that their performance across templates gets less consistent, this is really happening more at the edges. The middle 50% of templates for a player- templates where they perform better than they do on their worst 25% and worse than on their top 25%- has a much more consistent IQR (interquartile range- the difference in skill between their 75th percentile template and their 25th percentile template)- actually it's almost constant if you take out some players near the bottom (who'll have 0 inconsistency and IQR because they haven't played much). So as players get better, they get better by about the same amount across their average templates; they might be able to do some specific optimizations for some templates and lag behind a bit on some others, but there's still an empirically demonstrable "general skill" that better players have more of.



This means that a single-template ladder can be rather flawed because it's relatively easy to just adapt to that one template without having a corresponding improvement in general Warzone ability (at least on canonically strategic templates). But if you have a ladder like the MDL with considerable variation in templates, then you can get much more reliable results.

Edited 2/26/2018 18:38:28

"If you are below 400 you win 10 points and lose 5 points against an opponent of your rating. "

Ahh, my bad, was misremembering that before 400 it was +10-10, and after 400 was +5-10 as opposed to +10-5 and +5-5. You don't actually need to win 67% to rise past 400 then, just over 50%, though it needs to be 50% winrate vs people that also win at least 50%, not sure how the math boils down there.

I'd need to see what someone like mod gets from wins and loses from losses to better understand how it treats rating disparity in matchups.

I remember needing to win about 70% to it 500, but that was when mod was 700ish

"I think we should use it as the end-all-be-all instead of the 1v1 Ladder like Wally is attempting to do. "

Eh, I don't think anything should be an end-all-be-all. Each ladder/competition rates different things.

MDL is the best thing for determining overall warlight skill in multi-day games. It's a good way to determine the best warlight players, but not necessarily the best at specific templates due to sample size. I'd look to either QM or the 1v1 ladder as better solutions to the answer of, "Who are the top x players at strat 1v1?" for example. I'd pick QM over it for that question for any specific template most likely. At least, eventually, once most people have reached the point that playing more doesn't necessarily increase their ratings.

It's not that MDL is a better or worse answer, it's an answer to a different question.

Comes down to your interpretation of what it means to be the best Warzone player. I think that question itself (framed the way Wally has framed it in this post) can be meaningfully answered without having to ask "on which template?"

Top 5 is not that hard

1. Dead Piggy
2. Qi
3. Sze
4. Latnox
5. The Impaller.

Youre welcome

jesus wally, you made this thread yesterday and now it has 120 posts.

Wow, knyte, nice work! Especially the first post. What's the methodology you use for that (self-organising map? neural network?), and in what way is the data flawed? More than just sample-size issues?

I now see that my "There are... MDL templates that I haven't even played yet" comment was a massive understatement. Maybe I'll try it sooner rather than later after all....

But based on the templates I do know, it's hard to see the logical connection. In the top left you see Malvia and Battle Islands together, which makes a lot of sense as they are fairly standard and have a lot in common. However, with them sits Elitist Africa (hmm... maybe I'll join when they change the colour scheme) and Georgia Army Cap - two templates which are unusual in very different ways.

join now, veto africa

Didn't know you can veto templates. Soon....

It's a network graph (like a social network graph- code snippet for generating it can be found at https://gist.github.com/knyte/0936802177ae53d132dffbd540e9abf9#file-mdl_clustering-py). The connections between the nodes represent the strength of the correlation of player performance between them. I think the small sample size is the main issue.

Also can't really explain why it does that weird stuff with regard to mechanics. I don't have a lot of confidence in these exact clusters + due to the clustering algorithm used, some of these color-coded groups shouldn't really be together (as you can see with Unicorn Island being closer to most of the yellow than the greens). I can't really speculate as to why performance on Georgia Army Cap seems to have a comparatively strong correlation with performance on Battle Islands V- it could be the player pool, some weird similarities in the winning strategy, or just noise since even at this point in the MDL you can't really get accurate individual template-specific ratings for a most players. Since there's so much noise from the small sample set, it's much less likely for a specific conclusion (e.g., "Georgia Army Cap performance correlates closely with Greece LD performance") to be correct than for a general one (e.g., "The templates aren't all similar"). Don't really know how to solve this problem, even if I redid this with 3x the data, so for now I think we'll have to settle with broad-strokes interpretations of the data.

You can see some of the internals of this- a visual representation of player ratings on each template ("Ratings (Visual)") and of correlation of performance between templates ("Correlation (Visual)") at https://bit.ly/mdl-analysis.

I'll just put the images here for convenience, though. Here's the performance correlations between templates (note that even the really green tiles are relatively weak correlations by normal standards- that's partly due to noise and partly since no one template really predicts any other close to perfectly):



And player ratings on each template:







The ratings suffer from the same inaccuracy/noise issue as the correlations, so we can't *really* be sure that (for example) Timinator is as good as he appears to be on Battle Islands V, but we can be much more sure that Timinator was, as of the time of this analysis, pretty good at the template and better than the average player. Don't have enough data to be confident in narrow conclusions, so just have to settle for being more confident in broad-strokes ones.

Edited 2/26/2018 21:10:56

"maybe tom brady should try womens volleyball" was easily the best thing i've ever posted on the internet and it gets downvote hidden lol

I guess that tells us all we need to know about the rest of what you've said on the internet.

btw kicorse there are 7 vetos

#supportAfrica

fireice82 has been stalling full 3 days in our ladder game. as soon as i made this thread, he took his turns quickly. now he's stalling again.

scared he was going to get called out i guess.

there you go -- quit stalling.

Stalling by playing slow or stalling by playing slow in a lost game?