Helpstone

[Helpstone @helpstonex]

@Pier4r @TarjeiJS The problem with that is that it gets tougher the more players we have since the distribution will naturally have bigger tails. Probably difference to the average rating of the top 5/10/50 would be a good metric

[Helpstone @helpstonex]

@Saboo_Shubham_ Is it really Visual Studio and not vscode?

[Helpstone @helpstonex]

@moonlight_scum @BBCNews Isn't the black version the base version that's only one third as hot as the one referenced here?

[Helpstone @helpstonex]

@praisetheSuuN So hast du also Zugzwang gewonnen

[Helpstone @helpstonex]

@vyok_v @Ketimporta1427 @GMHikaru @loltyler1 Online rating systems are much better than the ELO system, which is outdated. The fact that the average online rating is higher is completely irrelevant and just due to scale and the initial rating. Rating systems are used to measure relative strength not absolute strength.

[Helpstone @helpstonex]

@bagdelensfordel @pawnalyze You could maybe tune your weights based on recent candidates data if you still have that data. I'm guessing tpr might be more informative than pre ratings in the later rounds of the tournament

[Helpstone @helpstonex]

@sp4rtaTwitch Tod, Hippo, Scars, Leon. Alles 2:0 wins

[Helpstone @helpstonex]

@Wc3Sunny @ixiXlorD @StarbuckSC2 @DonSl4sH @W3ChampionsTeam @Back2Warcraft but we do know that Starbuck "definitely got it right" and that S4lsh "totally gets his point"

[Helpstone @helpstonex]

@DonSl4sH @sp4rtaTwitch @EPICDUDESGAMING woher soll sp4rta denn wissen, dass das j in dudes stumm ist?

[Helpstone @helpstonex]

@rasbt Another idea would be to split the dataset in batches that are of nearly equal token length so that you only have to do minimal padding for each batch, but maybe the performance would suffer because the data isn't iid anymore

[Helpstone @helpstonex]

@rasbt I rarely see this done and I suspected it's because you maybe can't fully utilize the parallel processing speed of the dataloaders if you always have to tokenize between each batch

[Helpstone @helpstonex]

@Back2Warcraft It's ofc not the only factor and each factor applies differently to each tournament. Hitman playing, players slowly starting to develop new strategies (not rly the case for the invitational) and many more reasons also play a role and the new format is certainly not the only cause

[Warcraft3Info @Warcraft3Info]

By courtesy of @helpstonex, we are delighted to share his findings in the matchups' respective biases, based on data of the @W3ChampionsTeam season 12 and 13:

https://t.co/s98RaYe0nv

[Helpstone @helpstonex]

@ZsaszDf @Back2Warcraft Ofc this would most likely even out if the sample size is big enough but it's not.

You are basically assuming that a 2200 winning vs a 2000 player is just as meaningful as a 2100 winning vs another 2100.

[Helpstone @helpstonex]

@ZsaszDf @Back2Warcraft So what I illustrated with Focus translates to whole MMR ranges. E.g. maybe most NEs in that arbitrary range are at the higher end of that range. So it would seem like NE is doing super well WR wise but it's actually just because NE has the stronger players in that range.

[Helpstone @helpstonex]

@RehcraM_ @ScarsWc3 @JannesOfficial Maybe influence would have been the better word. I'm referring to blizzard often blindly following Remo's doc since it got the most views and they often don't invest time to check other sources