|
oov's record (from start) to his ELO peak is 77 wins - 20 losses (79.38%).
Bisu (after OSL loss to stork) to his current ELO peak is 61 wins - 17 losses (78.21%).
Bisu also had the benefit from being around 2130 when he started his current streak as opposed to oov who started with 2000 ELO via TLPD.
Thus, what inferences can we draw from this?
1. It's fairly reasonable to say that oov's "longer domination" of approximately the same win percentage (97 games vs. 78 games for Bisu so far) is what made up the extra ~130 or so point difference.
16 wins will make up 130 points (even with 3 losses) considering that I believe TLPD awards max 16 points (k value = 16) for beating opponents that are significantly above your level (if you were 2000 and they were 2100+). Points gained drops as your ELO increases.
2. Neither player is really "more dominant" to their opposition in regards to win percentage. Oov, on the other hand, WAS more dominant in respect to sustaining the win percentage longer.... so far.
-----------------------------------------------------------------
The system is slowly inflating to some extent, but it is unclear how much. The way you could look at it is see how many points have entered the system vs. how many have left (via people who retired).
For example, if there are two people who have retired one with 1900 points and one with 2200 points... the average is 2200+1900 - 2*2000 = 100 points have left the system = net deflation.
Whatever the case, the number is most likely VERY small because SC as a game hasn't had as many players entering and retiring as say chess or any other game that also uses ELO. With how many active players there are now as compared to people who retired, the system has probably only inflated maybe 5 points MAX to be honest. So saying that there's a HUGE inflation which is responsible for the higher ELO is fairly incorrect *at least* by normal deflation method.
The reason why Boxer and many of the older players have peak ELOs that are low is that
(1) not all of the games have been recorded that they played,
(2) sometimes they skipped out on certain tournaments in favor of others,
(3) MOST IMPORTANTLY, when the system starts the TRUE rating of the players is not what they are rated. At the beginning of the implementation of the database, each player starts with 2000 ELO points which is not accurate of their actual skill level.
For example (of 3), see the stats and corresponding ELO's of players on the map destination:
http://www.teamliquid.net/tlpd/maps/188_Destination/player_stats
JangBi, who is 12-4 on the map, is rated a paltry 2123 ELO which is not an accurate representation of his actual ELO over his games on all maps compared to others is 2282... a large margin. Thus, those starting out, aka the older players, have the disadvantage of having to have everyone increase their ratings over time as a group... and be dominant within that group... to see ratings increase.
-----------------------------------------------------------------
There are two fairly "accurate" methods then of seeing who is truly a more dominant player. I am not sure if this can be implemented with TLPD because I don't know how to go back in time... but maybe one of the mods can or something.
Let's look at oov's peak ELO of 2329. We can examine:
1. the average ELO of ALL active progamers during his peak.
2. the top 5 or maybe 10 players ELOs who are after him (and their overall average too.. for example, Bisu may have 3-4 strong challengers, but the rest are waaay behind their level).
These two stats would show (1) how dominant he is compared to the average player and (2) how strong he was to his top competition.
I would like to do this with Boxer, NaDa, iloveoov, sAviOr, FlaSh, Jaedong and Bisu... see how well they fare. My inclination is that Boxer and NaDa might be suprising with their dominance since they were the "earlier" Bonjwa's when the system was just coming into existence.
As far as oov and savior vs the 3 kings now... I am not sure what to expect. I do think that oov might be a bit stronger than his ELO indicates at least in regards towards his next top5/10 competitors, BUT I would speculate that he may have not been as dominant over the average player... in regards to his winning percentages being close to Flash/Jaedong/Bisu at peak levels.
|
Nice post eshlow, seems accurate and hopefully should help clear up some confusion and misconceptions about the system.
|
u need to do standard deviations, not averages
|
The whole point of the ELO is that inflation doesn't affect it.
You can add a million 1point ELO players over night and have bisu 1million - 0 them all and his ELO wouldn't budge.
|
On February 23 2009 17:04 fearus wrote:
The whole point of the ELO is that inflation doesn't affect it.
You can add a million 1point ELO players over night and have bisu 1million - 0 them all and his ELO wouldn't budge.
Uh, what does your second point have to do with your .. incorrect first point? If people started at 1 point you MIGHT have a point but they start at 2000.
ELO operates on a bell curve. It's simple really, you don't have that many people on the high end simply because if you lose with a high ELO you tank a lot of points. Everyone gets this i'm sure.
Starcraft has seen an increase of players and generally those players are better. What really happens in this situation is that the bell curve begins shifting to the right and in doing so inflates the middle point. When Nada was around he was the only person anywhere near 2250 ELO rating and only 1 of like 4 people above 2200 at the time. Currently we have quite a few more with 2 people within 25 points of 2300 and one person above 2300.
You don't require a large influx of players for an ELO to inflate, especially with a K rating of 16.
|
On February 23 2009 16:58 traced wrote:
u need to do standard deviations, not averages
Well, you need to find the average for the ratings...
But yes, a standard deviation would be the correct way to analyze *the difference* between the ratings.
|
On February 23 2009 17:13 Jayme wrote:Show nested quote +On February 23 2009 17:04 fearus wrote:
The whole point of the ELO is that inflation doesn't affect it.
You can add a million 1point ELO players over night and have bisu 1million - 0 them all and his ELO wouldn't budge. Uh, what does your second point have to do with your .. incorrect first point? If people started at 1 point you MIGHT have a point but they start at 2000. ELO operates on a bell curve. It's simple really, you don't have that many people on the high end simply because if you lose with a high ELO you tank a lot of points. Everyone gets this i'm sure. Starcraft has seen an increase of players and generally those players are better. What really happens in this situation is that the bell curve begins shifting to the right and in doing so inflates the middle point. When Nada was around he was the only person anywhere near 2250 ELO rating and only 1 of like 4 people above 2200 at the time. Currently we have quite a few more with 2 people within 25 points of 2300 and one person above 2300. You don't require a large influx of players for an ELO to inflate, especially with a K rating of 16.
Hey, I went to read a bit more on ELo and realised I was talking out of my ass. So thanks for correcting me up on that.
So elshow was absolutely correct in the way he said in order to view dominace we need to take into consideration "the median of players" and the relative ELo os number 2,3,4 etc.
|
On February 23 2009 17:34 eshlow wrote:Show nested quote +On February 23 2009 16:58 traced wrote:
u need to do standard deviations, not averages Well, you need to find the average for the ratings... But yes, a standard deviation would be the correct way to analyze *the difference* between the ratings.
i'm unsure what you mean by "ratings"
and yes, i was referring to comparison
|
So Bisu sits 3 points out, and he plays Movie tomorrow in set 1 of WL. Since I don't understand all of this math junk, can someone tell me where that would put him number wise? I know he'd obviously be ahead of Flash, but still don't know how far ahead he'd be, and I can't figure out this math stuff.
|
On February 24 2009 06:48 vx70GTOJudgexv wrote:
So Bisu sits 3 points out, and he plays Movie tomorrow in set 1 of WL. Since I don't understand all of this math junk, can someone tell me where that would put him number wise? I know he'd obviously be ahead of Flash, but still don't know how far ahead he'd be, and I can't figure out this math stuff.
Wouldn't be so sure there, Movie's ELO is only 2081..
|
On February 24 2009 05:50 traced wrote:Show nested quote +On February 23 2009 17:34 eshlow wrote:On February 23 2009 16:58 traced wrote:
u need to do standard deviations, not averages Well, you need to find the average for the ratings... But yes, a standard deviation would be the correct way to analyze *the difference* between the ratings. i'm unsure what you mean by "ratings" and yes, i was referring to comparison
To do a standard deviation you need to have a normalized point for the belle curve... which is the average ratings of all active players.
|
On February 24 2009 07:25 Rostam wrote:Show nested quote +On February 24 2009 06:48 vx70GTOJudgexv wrote:
So Bisu sits 3 points out, and he plays Movie tomorrow in set 1 of WL. Since I don't understand all of this math junk, can someone tell me where that would put him number wise? I know he'd obviously be ahead of Flash, but still don't know how far ahead he'd be, and I can't figure out this math stuff. Wouldn't be so sure there, Movie's ELO is only 2081..
He'd either tie or at best get 2 more than flash.
|
Bisu is going to all-kill anyway, so hes going to rape Flash's record.
|
Braavos36401 Posts
if Bisu wins against Movie, his rating goes up +7 to 2340, which is a new ELO record
if Bisu loses to Movie, his rating drops -33 to 2300 lol
|
Wow ... such a huge hit if he loses to Movie ... but no worries, he's going to All Kill again today ^^
|
United States12607 Posts
On February 25 2009 05:14 Hot_Bid wrote:
if Bisu wins against Movie, his rating goes up +7 to 2340, which is a new ELO record
if Bisu loses to Movie, his rating drops -33 to 2300 lol
Wow, I never realized dropping just one game to a low-ranked player could be so devastating. No wonder ELO peaks are tough to beat.
|
On February 25 2009 05:14 Hot_Bid wrote:if  Bisu wins against Movie, his rating goes up +7 to 2340, which is a new ELO record if Bisu loses to Movie, his rating drops -33 to 2300 lol
Well, then Bisu better win huh =]
|
Bisu #1 Forever <3
FAITH
PRIDE
SEXY
|
On February 25 2009 05:14 Hot_Bid wrote:
if Bisu wins against Movie, his rating goes up +7 to 2340, which is a new ELO record
if Bisu loses to Movie, his rating drops -33 to 2300 lol
Hot Bid,
Is there a way to go back in time with TLPD to analyze stats?
And thanks for that simulation.  Go Bisu!
|
Braavos36401 Posts
On February 25 2009 06:32 eshlow wrote:Show nested quote +On February 25 2009 05:14 Hot_Bid wrote:
if Bisu wins against Movie, his rating goes up +7 to 2340, which is a new ELO record
if Bisu loses to Movie, his rating drops -33 to 2300 lol Hot Bid, Is there a way to go back in time with TLPD to analyze stats? And thanks for that simulation. Go Bisu!
not sure, you'd have to ask Pop for that information.
TLPD can do a lot of things, but time travel might be beyond its powers.
|
|
|
|
|
|
EPT
