EPTData analysis on 8 million games - Page 4
| Forum Index > BW General |
|
Dakota_Fanning
Hungary2359 Posts
| ||
|
iloveav
Poland1479 Posts
| ||
|
sexyMIStrZZZ
Poland11 Posts
On October 13 2023 15:09 Dakota_Fanning wrote: I've already handed over a new dataset including 8.5 million games with MMR info included. Be patient until Kraekkling can analyze the new dataset and extract meaningful stats. How are you gonna know that data about 8.5m games? People change nicknames all the time and the MMR varies over time... Hard to imagine the MMR data to be any accurate. | ||
|
Dakota_Fanning
Hungary2359 Posts
On October 13 2023 17:13 sexyMIStrZZZ wrote: How are you gonna know that data about 8.5m games? People change nicknames all the time and the MMR varies over time... Hard to imagine the MMR data to be any accurate. The data is from Blizzard ladder, and the MMRs are from the time when the game was played. | ||
|
TMNT
3102 Posts
1st sample: Terran - 2300 - win 2nd sample : Protoss - 2100 - win .... 8000000th sample: Zerg 2000 - win Or you can only take the games where both players are in a same range of mmr, say 2100-2300, then 2300-2500,...and leave all the games that have a mmr discrepancy out. I think you could do both. Interesting to see if there is a big difference between the two methods. | ||
|
Kraekkling
580 Posts
- mmr difference between players < 200 (can be adjusted to be looser or tighter) - determine "match-mmr" by taking the mean of the mmr of both players - present results in "match-mmr"-brackets, e.g. 1800-2000, 1900-2100, 2000-2200,..., 2400++, etc I'll likely try out a few things to see where we run into problems due to low statistics. I'm looking for input on of how to group the games/players, feel free to suggest ideas. I've already handed over a new dataset including 8.5 million games with MMR info included. Be patient until Kraekkling can analyze the new dataset and extract meaningful stats. I'll probably get to it on the weekend. | ||
|
Kraekkling
580 Posts
On October 13 2023 16:25 iloveav wrote: Based on this data, 6 pool i op zvp. That's not how I would read it. What we see in the data just means that at this point in the game, it is much more likely for Zerg to win, because those are the games which end because the 6pool was successful. But games where the Protoss manages to defend will be longer and you can't pinpoint a single point in time where we expect Protoss to convert their advantage into a win. You're likely to instantly win games with a successful rush - but you don't instantly win games where you successfully defended a rush. We see this in all matchups - the race which has the "quickest rush" will win more games in the very early period of the game. | ||
|
TMNT
3102 Posts
Kraekkling, can you do something like "Race vulnerability vs mmr"? That is - you filter out all the games where the mmr difference is 300 or less, meaning you'll only work with games where there is a 300+ mmr diffrence (A) - you count the games where the loser is the one with higher mmr (B) - then (B)/(A) is your "vulnerability rate" of the race, which reflects how a supposedly better player can be beaten by a supposedly weaker player by cheese/luck/build order... Of course you'd do that for each mmr range to see when a race becomes more solid. From instinct we'd probably say Terran is the most vulnerable while Zerg is probably the most resistant from dying by cheese. And I guess at higher ranks when skill improves we'd see all 3 races become less vulnerable. But it'd be nice to see it represented with numbers. Also I'd expect to see B rank Protosses win the most vs players 300 mmr higher than them lol. | ||
|
WombaT
Northern Ireland26222 Posts
| ||
|
NotJumperer
United States1371 Posts
| ||
|
sexyMIStrZZZ
Poland11 Posts
On October 13 2023 17:23 Dakota_Fanning wrote: The data is from Blizzard ladder, and the MMRs are from the time when the game was played. I hope the data will be reliable and if it will then I'm pretty confident that Protoss at a high level will be performing the worst out of all 3 races. | ||
|
ZeroByte13
780 Posts
On October 13 2023 03:56 sexyMIStrZZZ wrote: Terran is so OP that in last 4 ASL tourneys top-8 werebecause Terran becomes so overpowered that it dominates every high level ladder/high level tournament and that is an issue of balance. ASL 13 - 3 T, 3 Z, 2 P ASL 14 - 3 T, 3 Z, 2 P ASL 15 - 2 T, 5 Z, 1 P ASL 16 - 3 T, 4 Z, 1 P Total - 11 T, 15 Z, 6 P Top-16 were ASL 13 - 4 T, 7 Z, 5 P ASL 14 - 6 T, 6 Z, 4 P ASL 15 - 6 T, 7 Z, 3 P ASL 16 - 6 T, 6 Z, 3 P Total - 22 T, 26 Z, 15 P Protoss suck the most at the very top level, yeah. But if anything - it's Zerg who's OP according to the ASL top-8 / top-16 statistics.  Of course, if you're looking only at top-1 or top-2... | ||
|
sexyMIStrZZZ
Poland11 Posts
| ||
|
Kraekkling
580 Posts
tesagi confirmed | ||
|
Crimson)S(hadow
Philippines594 Posts
| ||
IntoTheWow
is awesome32277 Posts
| ||
|
Cygnus
United States847 Posts
| ||
|
TMNT
3102 Posts
| ||
|
Rainalcar
Croatia422 Posts
On October 16 2023 03:00 Kraekkling wrote: Added MMR data. Also added overall win rate numbers. + Show Spoiler + tesagi confirmed Completely expected. I am a bit surprised to see TvZ slightly more favoured to T than ZvP to Z though. One thing is beyond doubt though - P has it by far the hardest at pro level. If the game were updated, this would have been patched pronto. | ||
|
sexyMIStrZZZ
Poland11 Posts
a) Terrans train harder, Snow and Dewalt don't. Don't ask, they just don't. Just check the ladder, terrans have the most games, it's not like you can just delete an account and create a new one within 20 seconds. b) It's about the maps, the maps are the reason Terrans win c) Have you seen how many games do Protosses win at 5 minute mark? That's absolutely disgusting d) But when Jyj won ASL he played against Protoss only once and he lost that one game. Check-mate Protossists! e) Have you seen any Terran beat Protoss that is 400 MMR higher than them? f) Protoss has invisible units? That can't be fair, it's fucking bullshit g) I believed that P>T for 20 years, so there's nothing that can convince me, you're just wrong h) Someone messed up the data and its actually the other way around i) Did you see Snow beat Flash? Well, you just gotta train harder, if he could do that then you can do it too (completely forgetting Snow's performance in the last ASL, that didn't happen at all guys) Just remember - never talk about facts, at least not when they undoubtebly prove that Terran is overpowered at a high enough level (which here is already above 2000 MMR and probably even slightly below 2k if you check the stats closely). | ||
| ||
