On February 04 2022 11:10 honorablemacroterran wrote: I think it's amazing the amount of copium people have to be on to say that Group B and Group D are comparably competitive.
Eh... per Aligulac (as of 02/04)
Group D – Maru (3354), Reynor (3193), Clem (3099), Zoun (2739), Scarlett (2657) Group B – Serral (3325), Dark (3191), ShoWTimE (3021), TIME (2743), SpeCial (2613)
Average Elo of 3008 and 2978, respectively.
Group C – Zest (3116), HeRoMaRinE (2963), Trap (2880), Bunny (2868), Lambo (2798) Group A – Rogue (3131), Cure (3043), Solar (3030), Dream (2732), Neeb (2665)
Average Elo of 2925 and 2920, respectively.
Not sure what 'copium' is, but I think it's fair to say that while Group D is definitely the hardest, Group B is closer to being as hard as Group D than Groups C and A are to being as hard as Group B. Depends on what you mean by competitive though, I guess...
Aligulac is shit.
Very incisive... I reckon you couldn't out-predict it though, over the course of, say, even one whole Liquibet season. It's far from perfect, to be sure. Sheer volume of games played, for instance, plays a larger role in players' ratings than perhaps it ought, which can lead a player like ByuN to have a rating that is somewhat inflated relative to that of a player like Rogue, at a given point in time. And recency plays no direct role, so if Neeb, say, has barely practiced in the last few months, he may be wildly overrated. Offline/online and premier/minor classification play no role either, although frankly, it's far from clear that they ought to, so I don't view that as a valid criticism in itself.
On the whole, albeit with several qualifications, I'd say its Elo rating system is a reasonably good barometer of players' relative odds of winning against one another on any given day. I'd venture to say that a very knowledgable (and uncommonly numerate) person can make use of its inference algorithm to augment his/her own judgements and out-predict the algorithm working by itself, but I very much doubt that anyone, using only his/her own knowledge and intuition, could out-predict it across, say, a sample of 10,000 – or even 1,000 – tournament matches, unless he/she went to the substantial trouble of tapping Aligulac's vast database of games to develop a competing inference algorithm (with somewhat different parameters) with which to augment his/her own apish intuitions.
On February 05 2022 13:56 honorablemacroterran wrote:
On February 05 2022 13:00 kochanfe wrote:
On February 04 2022 11:10 honorablemacroterran wrote: I think it's amazing the amount of copium people have to be on to say that Group B and Group D are comparably competitive.
Eh... per Aligulac (as of 02/04)
Group D – Maru (3354), Reynor (3193), Clem (3099), Zoun (2739), Scarlett (2657) Group B – Serral (3325), Dark (3191), ShoWTimE (3021), TIME (2743), SpeCial (2613)
Average Elo of 3008 and 2978, respectively.
Group C – Zest (3116), HeRoMaRinE (2963), Trap (2880), Bunny (2868), Lambo (2798) Group A – Rogue (3131), Cure (3043), Solar (3030), Dream (2732), Neeb (2665)
Average Elo of 2925 and 2920, respectively.
Not sure what 'copium' is, but I think it's fair to say that while Group D is definitely the hardest, Group B is closer to being as hard as Group D than Groups C and A are to being as hard as Group B. Depends on what you mean by competitive though, I guess...
Aligulac is shit.
Very incisive... I reckon you couldn't out-predict it though, over the course of, say, even one whole Liquibet season. It's far from perfect, to be sure. Sheer volume of games played, for instance, plays a larger role in players' ratings than perhaps it ought, which can lead a player like ByuN to have a rating that is somewhat inflated relative to that of a player like Rogue, at a given point in time. And recency plays no direct role, so if Neeb, say, has barely practiced in the last few months, he may be wildly overrated. Offline/online and premier/minor classification play no role either, although frankly, it's far from clear that they ought to, so I don't view that as a valid criticism in itself.
On the whole, albeit with several qualifications, I'd say its Elo rating system is a reasonably good barometer of players' relative odds of winning against one another on any given day. I'd venture to say that a very knowledgable (and uncommonly numerate) person can make use of its inference algorithm to augment his/her own judgements and out-predict the algorithm working by itself, but I very much doubt that anyone, using only his/her own knowledge and intuition, could out-predict it across, say, a sample of 10,000 – or even 1,000 – tournament matches, unless he/she went to the substantial trouble of tapping Aligulac's vast database of games to develop a competing inference algorithm (with somewhat different parameters) with which to augment his/her own apish intuitions.
I believe it systematically overrates European players relative to Koreans as I explained in the edited version of that post. Rating Reynor over Rogue is pretty apish though, as is believing the computer is always right.
On February 04 2022 11:10 honorablemacroterran wrote: I think it's amazing the amount of copium people have to be on to say that Group B and Group D are comparably competitive.
Eh... per Aligulac (as of 02/04)
Group D – Maru (3354), Reynor (3193), Clem (3099), Zoun (2739), Scarlett (2657) Group B – Serral (3325), Dark (3191), ShoWTimE (3021), TIME (2743), SpeCial (2613)
Average Elo of 3008 and 2978, respectively.
Group C – Zest (3116), HeRoMaRinE (2963), Trap (2880), Bunny (2868), Lambo (2798) Group A – Rogue (3131), Cure (3043), Solar (3030), Dream (2732), Neeb (2665)
Average Elo of 2925 and 2920, respectively.
Not sure what 'copium' is, but I think it's fair to say that while Group D is definitely the hardest, Group B is closer to being as hard as Group D than Groups C and A are to being as hard as Group B. Depends on what you mean by competitive though, I guess...
Aligulac is shit. It predicted that MaxPax was favored over PartinG 5 months ago when I was playing with it, and when PartinG was a Ro4 GSL player and MaxPax wasn't as good as he is now. You should go by the points that were created to qualify people for this event. The ratings in your own post discredit Aligulac as far as I'm concerned without going after any individual players when it's not their fault. Suffice to say, it systematically overrates Europeans relative to Koreans.
I'll grant you that the phenomenon of Koreans' regularly playing against Europeans on the European server in the ESL Open Cups over the past couple of years has almost certainly created some measure of rating inflation among European players (as, perhaps, has the more general abundance of cross-server online tournament play since early 2020), but I don't think the effect is all that large, to be honest...
On February 05 2022 13:56 honorablemacroterran wrote:
On February 05 2022 13:00 kochanfe wrote:
On February 04 2022 11:10 honorablemacroterran wrote: I think it's amazing the amount of copium people have to be on to say that Group B and Group D are comparably competitive.
Eh... per Aligulac (as of 02/04)
Group D – Maru (3354), Reynor (3193), Clem (3099), Zoun (2739), Scarlett (2657) Group B – Serral (3325), Dark (3191), ShoWTimE (3021), TIME (2743), SpeCial (2613)
Average Elo of 3008 and 2978, respectively.
Group C – Zest (3116), HeRoMaRinE (2963), Trap (2880), Bunny (2868), Lambo (2798) Group A – Rogue (3131), Cure (3043), Solar (3030), Dream (2732), Neeb (2665)
Average Elo of 2925 and 2920, respectively.
Not sure what 'copium' is, but I think it's fair to say that while Group D is definitely the hardest, Group B is closer to being as hard as Group D than Groups C and A are to being as hard as Group B. Depends on what you mean by competitive though, I guess...
Aligulac is shit.
Very incisive... I reckon you couldn't out-predict it though, over the course of, say, even one whole Liquibet season. It's far from perfect, to be sure. Sheer volume of games played, for instance, plays a larger role in players' ratings than perhaps it ought, which can lead a player like ByuN to have a rating that is somewhat inflated relative to that of a player like Rogue, at a given point in time. And recency plays no direct role, so if Neeb, say, has barely practiced in the last few months, he may be wildly overrated. Offline/online and premier/minor classification play no role either, although frankly, it's far from clear that they ought to, so I don't view that as a valid criticism in itself.
On the whole, albeit with several qualifications, I'd say its Elo rating system is a reasonably good barometer of players' relative odds of winning against one another on any given day. I'd venture to say that a very knowledgable (and uncommonly numerate) person can make use of its inference algorithm to augment his/her own judgements and out-predict the algorithm working by itself, but I very much doubt that anyone, using only his/her own knowledge and intuition, could out-predict it across, say, a sample of 10,000 – or even 1,000 – tournament matches, unless he/she went to the substantial trouble of tapping Aligulac's vast database of games to develop a competing inference algorithm (with somewhat different parameters) with which to augment his/her own apish intuitions.
I believe it systematically overrates European players relative to Koreans as I explained in the edited version of that post. Rating Reynor over Rogue is pretty apish though, as is believing the computer is always right.
The computer's not always right, but even relatively simple algorithms, like Aligulac's inference algorithm, can consistently out-predict even the most knowledgable human experts working alone – that is to say without the aid of a mathematical model to serve as a guide for their fallible intuitions...
Going against the grain here and saying its Group A. Can't get a much tougher Zerg draw with both Rogue and Solar. Dream doesn't get enough credit. He's in the second tier of Korean terrans but as HeroMarine said it, he's "extremely scary." Cure is arguably the best Terran in the world after Maru. Neeb is always a wild card because we don't see him in minor tournaments or Open Cups but has the ability to show some good games in this group.
On February 05 2022 13:56 honorablemacroterran wrote:
On February 05 2022 13:00 kochanfe wrote:
On February 04 2022 11:10 honorablemacroterran wrote: I think it's amazing the amount of copium people have to be on to say that Group B and Group D are comparably competitive.
Eh... per Aligulac (as of 02/04)
Group D – Maru (3354), Reynor (3193), Clem (3099), Zoun (2739), Scarlett (2657) Group B – Serral (3325), Dark (3191), ShoWTimE (3021), TIME (2743), SpeCial (2613)
Average Elo of 3008 and 2978, respectively.
Group C – Zest (3116), HeRoMaRinE (2963), Trap (2880), Bunny (2868), Lambo (2798) Group A – Rogue (3131), Cure (3043), Solar (3030), Dream (2732), Neeb (2665)
Average Elo of 2925 and 2920, respectively.
Not sure what 'copium' is, but I think it's fair to say that while Group D is definitely the hardest, Group B is closer to being as hard as Group D than Groups C and A are to being as hard as Group B. Depends on what you mean by competitive though, I guess...
Aligulac is shit.
Very incisive... I reckon you couldn't out-predict it though, over the course of, say, even one whole Liquibet season. It's far from perfect, to be sure. Sheer volume of games played, for instance, plays a larger role in players' ratings than perhaps it ought, which can lead a player like ByuN to have a rating that is somewhat inflated relative to that of a player like Rogue, at a given point in time. And recency plays no direct role, so if Neeb, say, has barely practiced in the last few months, he may be wildly overrated. Offline/online and premier/minor classification play no role either, although frankly, it's far from clear that they ought to, so I don't view that as a valid criticism in itself.
On the whole, albeit with several qualifications, I'd say its Elo rating system is a reasonably good barometer of players' relative odds of winning against one another on any given day. I'd venture to say that a very knowledgable (and uncommonly numerate) person can make use of its inference algorithm to augment his/her own judgements and out-predict the algorithm working by itself, but I very much doubt that anyone, using only his/her own knowledge and intuition, could out-predict it across, say, a sample of 10,000 – or even 1,000 – tournament matches, unless he/she went to the substantial trouble of tapping Aligulac's vast database of games to develop a competing inference algorithm (with somewhat different parameters) with which to augment his/her own apish intuitions.
This seems right, but I'm curious if Aligulac weights performance in high-stakes tourneys the same as, let's say, the weekly cups. Same question about online v. offline. Probably a decent enough metric overall, but there's no denying that players like Rogue have historically been underrated when it comes to Aligulac rating serving as a proxy for the likelihood of a successful performance in premier tournaments. By contrast, there was a stretch of time when Cure was the #1 rated player in Aligulac, and frankly I think everyone knew there were many others at that time who were better positioned to take the trophy in any given premier tournament. EPT points I think do a better job in this regard, even if it's higher variance given the fairly low number of data points. Aligulac may be better when it comes to mid-tier players, since it really only takes one good run or bombing out of a couple of tourneys in a row to throw things off. But I look at the EPT point distribution for the top players, and it feels pretty close to what it should be.
Most importantly, I believe Aligulac is a heavily volunteer-driven effort and no system is perfect, so I'm thankful there are folks out there willing to nerd out on this for our benefit.
On February 05 2022 13:56 honorablemacroterran wrote:
On February 05 2022 13:00 kochanfe wrote:
On February 04 2022 11:10 honorablemacroterran wrote: I think it's amazing the amount of copium people have to be on to say that Group B and Group D are comparably competitive.
Eh... per Aligulac (as of 02/04)
Group D – Maru (3354), Reynor (3193), Clem (3099), Zoun (2739), Scarlett (2657) Group B – Serral (3325), Dark (3191), ShoWTimE (3021), TIME (2743), SpeCial (2613)
Average Elo of 3008 and 2978, respectively.
Group C – Zest (3116), HeRoMaRinE (2963), Trap (2880), Bunny (2868), Lambo (2798) Group A – Rogue (3131), Cure (3043), Solar (3030), Dream (2732), Neeb (2665)
Average Elo of 2925 and 2920, respectively.
Not sure what 'copium' is, but I think it's fair to say that while Group D is definitely the hardest, Group B is closer to being as hard as Group D than Groups C and A are to being as hard as Group B. Depends on what you mean by competitive though, I guess...
Aligulac is shit.
Very incisive... I reckon you couldn't out-predict it though, over the course of, say, even one whole Liquibet season. It's far from perfect, to be sure. Sheer volume of games played, for instance, plays a larger role in players' ratings than perhaps it ought, which can lead a player like ByuN to have a rating that is somewhat inflated relative to that of a player like Rogue, at a given point in time. And recency plays no direct role, so if Neeb, say, has barely practiced in the last few months, he may be wildly overrated. Offline/online and premier/minor classification play no role either, although frankly, it's far from clear that they ought to, so I don't view that as a valid criticism in itself.
On the whole, albeit with several qualifications, I'd say its Elo rating system is a reasonably good barometer of players' relative odds of winning against one another on any given day. I'd venture to say that a very knowledgable (and uncommonly numerate) person can make use of its inference algorithm to augment his/her own judgements and out-predict the algorithm working by itself, but I very much doubt that anyone, using only his/her own knowledge and intuition, could out-predict it across, say, a sample of 10,000 – or even 1,000 – tournament matches, unless he/she went to the substantial trouble of tapping Aligulac's vast database of games to develop a competing inference algorithm (with somewhat different parameters) with which to augment his/her own apish intuitions.
This seems right, but I'm curious if Aligulac weights performance in high-stakes tourneys the same as, let's say, the weekly cups. Same question about online v. offline. Probably a decent enough metric overall, but there's no denying that players like Rogue have historically been underrated when it comes to Aligulac rating serving as a proxy for the likelihood of a successful performance in premier tournaments. By contrast, there was a stretch of time when Cure was the #1 rated player in Aligulac, and frankly I think everyone knew there were many others at that time who were better positioned to take the trophy in any given premier tournament. EPT points I think do a better job in this regard, even if it's higher variance given the fairly low number of data points. Aligulac may be better when it comes to mid-tier players, since it really only takes one good run or bombing out of a couple of tourneys in a row to throw things off. But I look at the EPT point distribution for the top players, and it feels pretty close to what it should be.
Most importantly, I believe Aligulac is a heavily volunteer-driven effort and no system is perfect, so I'm thankful there are folks out there willing to nerd out on this for our benefit.
Aligulac weights games from premier/minor events equally, as well as games from offline/online events equally, not discriminating in either case. That's why, as you point out, it tends to "overrate" a player like Cure, who plays (and often performs exceedingly well) in a large number of online events but doesn't often perform as well in premier offline events like GSL. Aligulac's inference algorithm, though, predicts the outcome of a generic match between two players, irrespective of whether it takes place online or offline or in a premier or minor event, so it will judge a match between, say, Rogue and Cure to have the same odds regardless of whether it takes place in an online qualifier (a context in which a human reasoner would rationally judge Cure to have a relatively better chance of winning) or a GSL Ro8 (a context in which a human reasoner would understandably judge Rogue to be a relatively heavier favorite).
Its inference algorithm's predictive validity might be improved if, when judging the odds on a given match, it gave relatively greater weight to previous results of like type (i.e. results from premier offline events if the match in question is taking place in such an event or else results from minor offline events if the match in question is of that sort). However, such a change would necessarily render it enormously more complex and might easily give rise to other kinds of systematic errors than the one its meant to correct for. It could equally be altered to give less weight to previous matches played across servers when tabulating the odds of an upcoming match played on the same server (let alone played offline).
Having said all of that though, it does generally do a pretty laudable job of predicting the results of a generic match between two opponents at a given point in time. Its relative simplicity and concomitant shortcomings are what render it able to be outshone by a tandem of itself (or some other hypothetical algorithm) and a human expert, but as I said, I highly doubt whether any single human expert could outdo it unaided over even a moderately large sample (such as 1,000 matches).
On February 05 2022 14:37 honorablemacroterran wrote:
On February 05 2022 14:35 kochanfe wrote:
On February 05 2022 13:56 honorablemacroterran wrote:
On February 05 2022 13:00 kochanfe wrote:
On February 04 2022 11:10 honorablemacroterran wrote: I think it's amazing the amount of copium people have to be on to say that Group B and Group D are comparably competitive.
Eh... per Aligulac (as of 02/04)
Group D – Maru (3354), Reynor (3193), Clem (3099), Zoun (2739), Scarlett (2657) Group B – Serral (3325), Dark (3191), ShoWTimE (3021), TIME (2743), SpeCial (2613)
Average Elo of 3008 and 2978, respectively.
Group C – Zest (3116), HeRoMaRinE (2963), Trap (2880), Bunny (2868), Lambo (2798) Group A – Rogue (3131), Cure (3043), Solar (3030), Dream (2732), Neeb (2665)
Average Elo of 2925 and 2920, respectively.
Not sure what 'copium' is, but I think it's fair to say that while Group D is definitely the hardest, Group B is closer to being as hard as Group D than Groups C and A are to being as hard as Group B. Depends on what you mean by competitive though, I guess...
Aligulac is shit.
Very incisive... I reckon you couldn't out-predict it though, over the course of, say, even one whole Liquibet season. It's far from perfect, to be sure. Sheer volume of games played, for instance, plays a larger role in players' ratings than perhaps it ought, which can lead a player like ByuN to have a rating that is somewhat inflated relative to that of a player like Rogue, at a given point in time. And recency plays no direct role, so if Neeb, say, has barely practiced in the last few months, he may be wildly overrated. Offline/online and premier/minor classification play no role either, although frankly, it's far from clear that they ought to, so I don't view that as a valid criticism in itself.
On the whole, albeit with several qualifications, I'd say its Elo rating system is a reasonably good barometer of players' relative odds of winning against one another on any given day. I'd venture to say that a very knowledgable (and uncommonly numerate) person can make use of its inference algorithm to augment his/her own judgements and out-predict the algorithm working by itself, but I very much doubt that anyone, using only his/her own knowledge and intuition, could out-predict it across, say, a sample of 10,000 – or even 1,000 – tournament matches, unless he/she went to the substantial trouble of tapping Aligulac's vast database of games to develop a competing inference algorithm (with somewhat different parameters) with which to augment his/her own apish intuitions.
I believe it systematically overrates European players relative to Koreans as I explained in the edited version of that post. Rating Reynor over Rogue is pretty apish though, as is believing the computer is always right.
The computer's not always right, but even relatively simple algorithms, like Aligulac's inference algorithm, can consistently out-predict even the most knowledgable human experts working alone – that is to say without the aid of a mathematical model to serve as a guide for their fallible intuitions...
Has anyone ever actually gone through liquibet predictions vs aligulac predictions and compared them over the course of a few 100 matches? It'd be a lot of work but it would be interesting using a point system to compare based on how strongly a match is called a certain way. With a comparison like that over a couple hundred series I'd be surprised if aligulac would come out significantly ahead or ahead at all. This comparison would even favor aligulac because liquibet is mostly only for GSL and I think the biggest problem with aligulac ratings/predictions is the cross server stuff.
I could be wrong and would love to see a true comparison but I don't view aligulac as being that good at predicting. It gets most matches right because most SC2 matches are pretty easy to predict nowadays. The top 10ish players generally don't lose anymore except to each other with very rare exceptions and that by itself is going to get a pretty high prediction rate.
On February 05 2022 14:37 honorablemacroterran wrote:
On February 05 2022 14:35 kochanfe wrote:
On February 05 2022 13:56 honorablemacroterran wrote:
On February 05 2022 13:00 kochanfe wrote:
On February 04 2022 11:10 honorablemacroterran wrote: I think it's amazing the amount of copium people have to be on to say that Group B and Group D are comparably competitive.
Eh... per Aligulac (as of 02/04)
Group D – Maru (3354), Reynor (3193), Clem (3099), Zoun (2739), Scarlett (2657) Group B – Serral (3325), Dark (3191), ShoWTimE (3021), TIME (2743), SpeCial (2613)
Average Elo of 3008 and 2978, respectively.
Group C – Zest (3116), HeRoMaRinE (2963), Trap (2880), Bunny (2868), Lambo (2798) Group A – Rogue (3131), Cure (3043), Solar (3030), Dream (2732), Neeb (2665)
Average Elo of 2925 and 2920, respectively.
Not sure what 'copium' is, but I think it's fair to say that while Group D is definitely the hardest, Group B is closer to being as hard as Group D than Groups C and A are to being as hard as Group B. Depends on what you mean by competitive though, I guess...
Aligulac is shit.
Very incisive... I reckon you couldn't out-predict it though, over the course of, say, even one whole Liquibet season. It's far from perfect, to be sure. Sheer volume of games played, for instance, plays a larger role in players' ratings than perhaps it ought, which can lead a player like ByuN to have a rating that is somewhat inflated relative to that of a player like Rogue, at a given point in time. And recency plays no direct role, so if Neeb, say, has barely practiced in the last few months, he may be wildly overrated. Offline/online and premier/minor classification play no role either, although frankly, it's far from clear that they ought to, so I don't view that as a valid criticism in itself.
On the whole, albeit with several qualifications, I'd say its Elo rating system is a reasonably good barometer of players' relative odds of winning against one another on any given day. I'd venture to say that a very knowledgable (and uncommonly numerate) person can make use of its inference algorithm to augment his/her own judgements and out-predict the algorithm working by itself, but I very much doubt that anyone, using only his/her own knowledge and intuition, could out-predict it across, say, a sample of 10,000 – or even 1,000 – tournament matches, unless he/she went to the substantial trouble of tapping Aligulac's vast database of games to develop a competing inference algorithm (with somewhat different parameters) with which to augment his/her own apish intuitions.
I believe it systematically overrates European players relative to Koreans as I explained in the edited version of that post. Rating Reynor over Rogue is pretty apish though, as is believing the computer is always right.
The computer's not always right, but even relatively simple algorithms, like Aligulac's inference algorithm, can consistently out-predict even the most knowledgable human experts working alone – that is to say without the aid of a mathematical model to serve as a guide for their fallible intuitions...
Has anyone ever actually gone through liquibet predictions vs aligulac predictions and compared them over the course of a few 100 matches? It'd be a lot of work but it would be interesting using a point system to compare based on how strongly a match is called a certain way. With a comparison like that over a couple hundred series I'd be surprised if aligulac would come out significantly ahead or ahead at all. This comparison would even favor aligulac because liquibet is mostly only for GSL and I think the biggest problem with aligulac ratings/predictions is the cross server stuff.
I could be wrong and would love to see a true comparison but I don't view aligulac as being that good at predicting. It gets most matches right because most SC2 matches are pretty easy to predict nowadays. The top 10ish players generally don't lose anymore except to each other with very rare exceptions and that by itself is going to get a pretty high prediction rate.
Well, it wouldn't be a very well-controlled experiment, considering the fact that many Liquibet participants (including, I suspect, most or all of the most successful ones) make a habit of glancing at Aligulac's inference algorithm prior to betting. I also would be mildly interested to see the results of such a comparison though. If I were to hazard a guess as to what it might show, I'd reckon that the bettors who cluster near the very top in any given season will substantially outperform the inference algorithm (examples of the human expert + algorithm tandems that I acknowledged could prove to be substantially more accurate than the algorithm alone), while many bettors in the upper-middle and middle of the distribution would perform about equally well as the algorithm (again, perhaps largely because most of them look at the algorithm and only end up being right about half of the time when they decide to deviate from its prediction), and the rest would be roundly beaten by the algorithm alone (either because they don't consult it at all or are wrong more often than not when they decide to deviate from its prediction).
It'd be mind-numbingly tedious to gather the necessary data though, because as far as I know its impossible to ask Aligulac to predict a match as though it were taking place on some date in the past rather than in the present. So you'd have to reverse-engineer the historical score, score vs. opposing race, form, form vs. opposing race, score vs. each other, and Elo vs opposing race of both players on the date of each match separately to generate predictions to compare against the betting.
It doesn't take a genius to outperform Aligulac when it's rating Reynor over Rogue and Clem. I don't know why anyone would consult it for anything. Everything it gets right should be fairly obvious to anyone who pays a modicum of attention, and lots of what it gets wrong should also be fairly obvious to anyone who pays a modicum of attention.
Maru best player in world right now, and GOAT without question. Serral, Rogue fans just go sit down please.
Clem and Reynor almost always deliver extremely high level games a lot of the time even if they don't win as much as they used too the skill level is undeniable.
Scarlett has been playing a lot better as of late and she is always scary in tournaments like these, when she gets a lot of time to plan for series she is a very scary player and it will not be the first time she causes major upsets.
Zoun is highly underrated player , he can play any style of protoss, which is important if you want to be relevant whatsoever as a protoss at the highest level, he also tries to innovate and do crazy builds, sometimes they fall flat and he can look pretty silly but he is a very creative and intelligent player.
This is why this is the Group of Death
As for who wins this group is have no idea, all I can say is Protoss never do well when you have to show consistency over many games to progress into a tournament so I'll be very surprised and happy if Zoun moves on because historically Protosses never do well in the tournaments that really matter.
I don't think there is any group of death as of now. Group of death has historically been a group of only strong players AND it is not obvious who will advance and who will be eliminated. Looking at each group and making a prediction that I don't think many will disagree with: A has Rogue advancing and Neeb getting eliminated. B has Serral and Dark advancing. C has Trap and Zest advancing. D has Maru and Reynor advancing. I can easily predict 2 players in each group. That's not something that should be possible in a group of death.
However, group of death has also historically been a double elimination group of 4. These are groups of 6, with round robin format. We also only know who 5 out of 6 players are in each group. 3 will advance and 3 will be eliminated. This is not a standard group of death question.
My thoughts of each group is the following: Maru, Clem, Reynor are 3 of the top tier players, with Maru>Reynor>Clem as an internal ranking. Both Scarlett and Zoun have shown that they can play evenly vs the best but their base line is lower. Group D is stacked with potential. In group C there is a trifecta of second tier players. Lambo, Bunny and Heromarine. They are very good, but not top tier. Group C has a good base. TIME, SpeCiaL and Showtime is about the same as the trifecta. The difference being the 2 Terran players in group C are varying wildly in performance. TIME can be top tier 1 week and mediocre the next while SpeCiaL is hit or miss in each match. Group B can have a lot of upsets, but they are clearly upsets. It is not expected that TIME 2-0 Serral and Dark but it could happen. Group B is volatile. Lastly is the group of 3 layers. Rogue is best. Dream, Solar and Cure are a step below. Neeb is good, but the others are better. Group A should have Rogue hiding strategies and advancing based off of his good foundation of skill. He might lose a few games because of hiding the strategies but what does it matter when he crushes people in the playoffs? Group A will be a contest between Dream, Cure and Solar. Which 2 of them will advance along Rogue?
If herO gets into group D, then I might call it a group of death. Otherwise I'd argue that there is no group of death because the level of skill within each group is too varied. The difference between best and worst players is too big for each group.
None of these groups is close to, say, GSL 2015 season 3. Group B had Maru, Classic, herO and sOs. Classic had just won GSL's sister tournament, SSL, and IEM Shenzen. He was rated as potentially the best player at the time. The meme of $o$ were going strong, he was the regular ace of Jinair in Proleague and his face was painted on a plane to honor his SC2 accomplishments. sOs went on to win at Blizzcon 2 months later. While Classic won SSL and IEM Shenzen, herO got to the semi finals of both those tournaments. herO did win the KeSPA cup that was run in the middle of the last SSL and GSL season, and herO also went on to win SSL season 3. This was peak herO. This was the Smiling Assassin. And lastly, we have Maru. The winner of the group. Maru-ders vs 3 of the top 5 Protoss players in the world. That was a group of death, one of many in code S history.
On February 09 2022 03:13 Vindicare605 wrote: LOL. Now Time and Cyan can't even compete so we're going to rotate some random alternate lower ranked player into Group B.
These groups are so lopsided. But it's ESL I don't know what I expected.
ESL has been relatively transparent about the process. These groups have been selected via tiers according to points. Top 4 players are distributed in the groups, then the next 4 players, then the next etc. Tier 1, score above 2669: Serral, Trap, Maru, Rogue. Tier 2, score between 2536 and 1976: Cure, Dark, Clem, Zest. Tier 3, score between 1914 and 1333: Reynor, TIME, Solar, Big Gabe. Tier 4, score between 1315 and 1237: Juanito, Neeb, Bunny, Scarlett. Tier 5, score below 1173: Dream, Lambo, Showtime, Zoun.
The lopsidedness of the groups is a consequence of player performance over the last year. People who have done well in tournaments get more points. Anyone unlucky enough to be eliminated early will get fewer points. The data points can be a bit few. They could have used more data points and just ripped off Aligulac, but that isn't really part of the entertainment that lures us viewers. Any system used will have flaws. ESL has been open about how the system works. What you expect should be exactly this. It is what we've had for a long time.
On February 09 2022 03:13 Vindicare605 wrote: LOL. Now Time and Cyan can't even compete so we're going to rotate some random alternate lower ranked player into Group B.
These groups are so lopsided. But it's ESL I don't know what I expected.
ESL has been relatively transparent about the process. These groups have been selected via tiers according to points. Top 4 players are distributed in the groups, then the next 4 players, then the next etc. Tier 1, score above 2669: Serral, Trap, Maru, Rogue. Tier 2, score between 2536 and 1976: Cure, Dark, Clem, Zest. Tier 3, score between 1914 and 1333: Reynor, TIME, Solar, Big Gabe. Tier 4, score between 1315 and 1237: Juanito, Neeb, Bunny, Scarlett. Tier 5, score below 1173: Dream, Lambo, Showtime, Zoun.
The lopsidedness of the groups is a consequence of player performance over the last year. People who have done well in tournaments get more points. Anyone unlucky enough to be eliminated early will get fewer points. The data points can be a bit few. They could have used more data points and just ripped off Aligulac, but that isn't really part of the entertainment that lures us viewers. Any system used will have flaws. ESL has been open about how the system works. What you expect should be exactly this. It is what we've had for a long time.
I think you're misunderstanding the issue. There is a constrained random draw process. The fact they are so lopsided is ostensibly because of the random draws from each tier and how they got assigned to each group. There are many ways the groups could have been drawn given the rules. They are as lopsided as they turned out to be 1% of the time.
On February 09 2022 03:13 Vindicare605 wrote: LOL. Now Time and Cyan can't even compete so we're going to rotate some random alternate lower ranked player into Group B.
These groups are so lopsided. But it's ESL I don't know what I expected.
ESL has been relatively transparent about the process. These groups have been selected via tiers according to points. Top 4 players are distributed in the groups, then the next 4 players, then the next etc. Tier 1, score above 2669: Serral, Trap, Maru, Rogue. Tier 2, score between 2536 and 1976: Cure, Dark, Clem, Zest. Tier 3, score between 1914 and 1333: Reynor, TIME, Solar, Big Gabe. Tier 4, score between 1315 and 1237: Juanito, Neeb, Bunny, Scarlett. Tier 5, score below 1173: Dream, Lambo, Showtime, Zoun.
The lopsidedness of the groups is a consequence of player performance over the last year. People who have done well in tournaments get more points. Anyone unlucky enough to be eliminated early will get fewer points. The data points can be a bit few. They could have used more data points and just ripped off Aligulac, but that isn't really part of the entertainment that lures us viewers. Any system used will have flaws. ESL has been open about how the system works. What you expect should be exactly this. It is what we've had for a long time.
I think you're misunderstanding the issue. There is a constrained random draw process. The fact they are so lopsided is ostensibly because of the random draws from each tier and how they got assigned to each group. There are many ways the groups could have been drawn given the rules. They are as lopsided as they turned out to be 1% of the time.
The point Drflip is making is that grouping players into tiers should not result in lopsided groups, even in the worst case. The reason why group D appears so much stronger is largely because Reynor (#9) and Scarlett (#13) play at a higher level than what their EPT rankings suggest. Any group assignment based on EPT points will suffer if the EPT points do not reflect player strength. Just as group assignment based on aligulac will seem unfair if aligulac ratings are not predictive.
IMO the "Group D is top 1% in lopsidedness" stat, which you keep repeating, is misleading. You're comparing group D to other groups from the tiered selection, and finding that it is more lopsided than 99% of them. But the groups resulting from tiered selection are pretty even - even the worst possible group has just 15% more EPT points than the average. (10k vs 8.5k)
What you should be comparing group D to are the groups assigned without the tier restriction. For instance, if Group E had Maru, Rogue, Trap, Serral, and Clem, for a total of 14225 EPT points. Or if Group F had Lambo, Showtime, Zoun, Dream, and Special, for a total of 5007 points. Now those are some lopsided groups. Not group D, which seems even by comparison.
EPT
