Well, I could…but I think I would lose some perspective in the data by doing so. The point of comparing “high skill” players to other “high skill” players is to simulate what a top table would look like in a tournament. So, a couple things:
The data is also restricted to dates after June 8, which is important for NEH. I could also open them up to dates prior to that, but then you start obscuring the effect that NEH has had (which I showed was significant in an earlier post)
I suppose I should have stated this earlier: in high skill vs. non-high skill players, it’s entirely possible that there is a different dynamic playing out in the global meta environment. We can’t forget that when looking at these analyses. This is a bit of a complex idea, so I can expand further if it doesn’t seem clear.
If I increase the data pool size by only requiring one side to be “high skill”, then you add in the complications of #2. Which is not to say that it is a bad thing, just that for this analysis I wanted to start with the “pure” data set of high skill vs. high skill, which would (in theory) reduce confounders and such.
So, long story short, perhaps that’s not a bad idea and I’ll do that on Monday. Just have to enter it with different mindset.
Ah I think you might have slightly misconstrued what I was suggesting; I might not have said it very clearly. I didn’t mean for you to consider high skill vs. non; I still want you to look at high vs. high but I was suggesting that some people might be high skill with one side but not the other.
At present you have your high skill threshold calculated over Corp and Runner games combined. If (for the sake of argument) the threshold happened to be 60, Player A with a Corp / Runner record of 62 / 58 (over an equal number of games on both sides) would make the cut while Player B with 63 / 56 would not, despite having a better Corp record than Player A.
So I’m suggesting that you apply the high skill cut separately for both sides, but still only consider games between players who are both “high skill” for their respective sides. I guess it all depends on what the point of the exercise is from your perspective. Your cut probably does “simulate what a top table would look like in a tournament” because the players with asymmetric (relative) win rates probably wouldn’t rise to the top, I’m just suggesting that we might get a larger data set and greater statistical significance if we didn’t ignore those players. From my point of view I’d rather know more reliably what the maximum potential of certain IDs are.
I don’t know if you’re applying any other cuts either? But in previous analyses there has been a cut on influence (to exclude games which are clearly being played with non-optimised decks). You might even consider discounting games with certain identities when calculating win rates for the player skill. For example, someone might be just missing the cut because of a couple of losses with The Professor in casual or test play, but if you excluded him from the calculation (because he’s clearly non-competitive at the moment) then you might increase your statistics further.
I don’t think theres’ necessarily any different information hiding in the extra statistics, but it would firm up what we do have. You could also consider a 2 sigma and 3 sigma cut for very high skill levels - then compare games between players in different tiers.
Ah. Ok. I see what you mean now. That is definitely doable.
I tried the 2 sigma already, and it really cut off a significant number of observations. I don’t remember exactly, but there were only a couple hundred records in that dataset, so I didn’t bother with it. Although it might be worth looking at again using the method you described.
I saw similar results with 2 sigma cutoffs on Glicko rating. The hard part of doing this is getting a dataset that’s representative of pretty good players while maintaining enough games to draw meaningful conclusions.
I’ll probably look at the matchup counts this weekend and see if any new IDs merit the same treatment I gave the core IDs plus Andy and Reina in my last set of articles.
wow Leela looking strong. I think good players have adjusted to her, but you can totally bomb out mid players with her.
wonder when HB will get a boost. I think BS does both Deadcoats and Glacier better than HB right now, and NEH does FA better
interesting that HB has the highest average points per game, but lowest win rate. from this I would infer that many HB players are playing Rush, but having trouble closing out the game.
idk if @db0 is busy or holidays stuff or what. The Source would generally be rezzed on octgn by now, seeing as it’s officially out today (they said 18th right? i know some shops already got it, unless TC was proxying Wednesday night)
What usually happens with HB is : they score their first 4/5 points (either by FA or in a remote) and then they’re unable to score either because their remote or the R&D is locked.
I’ve actually expressed some interest in how Slaghund gets the data streamed, since I have an 8TB NAS at home and could throw it all in MySQL or MongoDB and then query it on a whim whenever I want. But I realize that’s kind of greedy, so I haven’t pestered db0 about it.
But I realize that’s kind of greedy, so I haven’t pestered db0 about it.