|
Hey,
On May 24 2013 06:36 Grayson Carlyle wrote:
In other words, if a player with a rating of 1900 goes 1-0 against 4 players whose average rating is 1700, is it the same as going 4-0 against a single player in a single series?
Yes, this is the same.
On May 24 2013 06:36 Grayson Carlyle wrote:
The former's median outcome from prediction, on a game per game basis would be the sum of the expected median outcomes of each individual one: 4-0. Whereas the expected median outcome of the latter is 4-3.
Ah, the median predicted outcome (MPO) you see on the prediction pages are not the same as the expected outcome (EO) you see on the adjustment overview. The differences are:
- MPO is always integral, while EO are real numbers.
- When calculating MPO, we don't know how long the match will go. It's a measure of the closeness of the players, not a good estimate of the length of the match (in fact, it will overestimate match lengths consistently). The EO is expected score given the total number of games in the match.
So in your example, if the 1900 guy 4-0's the 1700 guy, that is four games. The EO can then be something like 2.8-1.2, or some other numbers who sum up to four. Of course these are impossible scores to actually get in a Bo7, even if we were to round them off, but the match format is inconsequential to the rating calculator.
So the EO is actual expected score given the games that were actually played (and is comparable to the same concept in Elo ratings), while the MPO is just "our best guess" at the actually possible outcome that would be least surprising.
Hope that clarified things.
|
That does clarify it a bit, but it doesn't address the issue that the expected outcome of a Bo1 is always on average 50% away from actual possible outcomes, so playing Bo1s still has a larger impact on a player's rating (both up and down) than longer series, whereas longer series are actually a better indication of a player's true rating, but they will inevitably have a lesser effect on the accuracy.
|
On May 24 2013 23:13 Grayson Carlyle wrote:
playing Bo1s still has a larger impact on a player's rating (both up and down) than longer series
I don't know why you say that, I told you that the distribution of the games into matches has no effect.
Playing four Bo1s against the same person and winning them all 1-0 has exactly the same effect as playing one Bo7 and winning it 4-0.
Playing one Bo1 and winning it 1-0 will gain you about one quarter of the points as playing one Bo7 and winning it 4-0. (Because in the first scenario there are only a quarter of the games).
In a longer match you can get closer to the expected outcome overall, but of course a longer match means more games, so your rating will change that much more.
|
Grayson: If you take several individual BO1s together, the average of the *absolute value* of the difference between the expected and actual outcome is 0.5. But you should not be ignoring the sign of the difference. Some will be wins and some losses.
|
On May 24 2013 23:47 TheBB wrote:
Playing four Bo1s against the same person and winning them all 1-0 has exactly the same effect as playing one Bo7 and winning it 4-0.
In a longer match you can get closer to the expected outcome overall, but of course a longer match means more games, so your rating will change that much more.
That's my point though, is that playing 4 Bo1s should not have the same effect since it is impossible for them to be anywhere close to the expected outcome except in a situation where a player is highly favoured over the other. The possible outcomes of 4 Bo1s against different people is far more limited and doesn't give nearly as much useful, predictive power as someone winning a Bo7 4-0. The latter is very significant for determining skill, whereas the Bo1s are not significant. However, they both have the same effect on rating.
|
Sorry, but I really can't understand why it's less significant for someone to win four Bo1s in a row than to win a Bo7 4-0, since they are the exact same thing.
If the expected outcome of a single game is 0.5-0.5, then after four Bo1s it's possible to be off by 2, 1, or 0, and after a Bo7 it's possible to be off by 2, 1.5, 1, 0.5 or 0. It provides finer resolution but the extremes are identical.
|
If you 4-0 someone, why should it matter if it was part of a BO7 (cut short because you already won) or 4 individual BO1s? Don't they both show exactly the same information? They played 4 times, and one player won all 4. Why should it matter if it was part of a match or not? If they played 3 more games, the previous 4-0 result should have no impact on the results of the next games.
If you are talking about psychological effects, yes those are ignored by pretty much all rating systems. There are all kinds of things outside of the game that are ignored -- like how much sleep the player got before the match etc.
==============
ETA an example conversation:
KD: Hey aligulac, TerranOP beat ZergSux 4-0. What are their new ratings?
Aligulac: KD, first you need to tell me, was the 4-0 part of a BO7? Or Two BO3s? Or 4 BO1s? Or maybe it's a BO101, but they had to stop early?
KD: I don't know, I just know they played 4 games and TerranOP won them all.
Aligulac: Sorry I can't update their ratings without knowing what the match format was.
This simply isn't the way rating systems work. They always assume each game is an independent event.
|
Different subject: I saw the graph of points transfer from foreigners to Koreans. You could go back to the beginning and increase the initial seeded ratings of Koreans until the points transfer is near zero.
|
On May 25 2013 02:15 KillerDucky wrote:
Different subject: I saw the graph of points transfer from foreigners to Koreans. You could go back to the beginning and increase the initial seeded ratings of Koreans until the points transfer is near zero.
Yeah, I've thought about doing exactly this. I'm sure it would help a lot with these perceived problems.
|
The most significant improvements made over the last month:
Stories: yellow dots on players' rating charts to show important career events. Also red dots for team switches.
Added comparison feature to help compare players directly against each other.
Improved error/success/warning/info messages for every page.
Player match history filtering significantly improved.
Condensed the small events lists.
Predictions are now default best of 1.
When searching for events, you can now use quotes to improve results (code S used to match code A because the S matches the S in GSL, now you can write "code s" to get what you want).
URL shortener function on every page (button on the bottom.)
|
Also
I've been promising a change in the rating algorithm for a while now, and finally it's here. The changes are numerous, but the most significant are:
Koreans now start with 1200 points, not 1000. This might be a bit controversial, but it goes a long way to fixing the rating discrepancy between the Korean and foreign scenes, and it is justified by the numbers.
Offline games are now weighted heavier than online games by a factor of 1.5. This has sped up the point transfer from ESF to KeSPA.
There are no longer any "fake games" in added to avoid infinities.
I won't promise that this is the final version, however!
|
A very small mistake, but in the OP, there is a milestone set in 2014 (rather than obviously 2013)
Apart from that, thanks for your job. It has de facto replaced TLPD for me. That's a damn awesome website.
|
2014-04-20: Links to Aligulac player profiles on Liquipedia.
2014 already? O.O
|
I guess now is as good a time as any to reveal that I'm the next doctor.
|
France15997 Posts
You killed the ForGGeddon, I will never forgive you. 
|
Scotland380 Posts
You're not really using statistics? you traveled to the future and know all the results?
|
On June 02 2013 06:26 Wodger wrote:
You're not really using statistics? you traveled to the future and know all the results?
* Doom zoom*
|
The recent model of basing the calculations seems to be the move in the right direction. I'm pleased with the increase to Korean numbers, and changing online tournaments to weigh less than offline is just perfect. Now when I look at the rankings, it seems reasonable and accurate because the inflation from online games is gone and KeSPA is better represented.
|
Love the new update. I think the offline factor is absolutely the right thing to do. People remember champions for offline tournaments, not online ones!
The Hall of Fame now much more resembles my memory of who I thought were dominant. Thanks for the hard work!
|
On June 02 2013 06:40 edwahn wrote:
Love the new update. I think the offline factor is absolutely the right thing to do. People remember champions for offline tournaments, not online ones!
The Hall of Fame now much more resembles my memory of who I thought were dominant. Thanks for the hard work!
The HoF wasn't really taken that much into consideration, we all agreed that it needed som rework. I guess this update had a good side effect on that as well 
|
|
|
|
|
|
EPT
