Home » Forum Home » General

Topic: provisional rating formula
Replies: 10   Views: 65,470   Pages: 1   Last Post: Oct 14, 2017, 1:10 AM by: watsu

Search Forum

Back to Topic List Topics: [ Previous | Next ]
Replies: 10   Views: 65,470   Pages: 1  
dmitriking

Posts: 375
Registered: Dec 16, 2001
Age: 40
provisional rating formula
Posted: Oct 11, 2017, 7:08 PM

the Provisional rating formula is listed as follows:

rating = (value + (rating * total1)) / total2


That equation has "rating" on both sides sides. I assume that the one on the left refers to "new rating" and the one on the right refers to "old rating"?

That said, I found that the change in my ratings did not match what I calculated it should be, using the above assumption.

If anyone can please shed some light on this, please do.


rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Re: provisional rating formula
Posted: Oct 11, 2017, 10:43 PM

Can you provide the numbers that didn't match what you expected?

dmitriking

Posts: 375
Registered: Dec 16, 2001
Age: 40
Re: provisional rating formula
Posted: Oct 13, 2017, 7:25 AM

Yes. on 9/18/17, I won a connect6 match against Scott. I was provisional and he was established. I had completed 2 matches previously and this was my 3rd. My rating was 2018 and Scott's was 1363.

According to my math, my new rating should have gone from 2018 to 1978.83, but instead of went to 2046.

As it happens, 2045.50 is the exact answer I get if I were to set e equal to w in the calculation. However, according to the instructions, e is supposed to equal w only if the opponent is also provisional, which my opponent was not. If my opponent is established, e is supposed to equal zero, if I'm understanding this correctly:


value = ( r1 + r2 ) / 2 + w * 200 + e * 200

Where r1 and r2 are the same as defined above.
w is 1 for a win and -1 for a loss.
e is 0 if your opponent is established, otherwise it equals w.

So, if the opponent is established, then e equals 0 and the 200e term drops out, leaving

value = (( r1 + r2 ) / 2 ) + 200

(I've changed 200w to simply 200 because I'm assuming w to be 1 that way I can eliminate a variable.

I then used the following formula for my expected new rating, substituting the above formula in place of "value", and also changing "total2" to "1+total 1," since that will always be the case and it further eliminates a variable.

Thus, for the new rating formula for provisional versus established, I have the following:



New rating = [ ((r1 + r2)/2 ) + 200 + (r1T1) ] / (1 + T1)

If the opponent is also provisional, then e equals w, and the 200w term becomes 400w, so we would just substitute 400 into the above equation in place of 200.

rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Re: provisional rating formula
Posted: Oct 13, 2017, 10:13 AM

> e is 0 if your opponent is established, otherwise it
> equals w.

You are correct in observing that the code is contradicting the documentation, and sets e to w if the opponent is established.

It seems the value if e is designed to give wins and losses an extra boost for provisional players, and the code says to give that extra boost if the opponent is established.

It makes sense to me to interpret it that way, provisional players get an extra bonus if they perform unexpectedly well against an established player. The risk of being rewarded extra is matched by the risk of losing double in case of a loss.

Now: is there an alternative interpretation as to why the formula should be as described in the help pages and not the code?

dmitriking

Posts: 375
Registered: Dec 16, 2001
Age: 40
Re: provisional rating formula
Posted: Oct 13, 2017, 7:25 PM

It makes more sense to me the way it's coded, but I haven't really given it much thought and I'd be curious what others think. Which one is the official actual rule - the written one or the coded one? If the official rule as used in chess and other major events is the rule as written, I assume there's a good reason.

rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Re: provisional rating formula
Posted: Oct 13, 2017, 7:31 PM

The rule that's in effect is the one that's coded, but I can only guess (see previous message) as to why that is, that implies that I am guessing the rule in the help pages is a typo.

Reasons to keep the coded rule and not the written one include that changing the coded rule
- makes previously calculated ratings a bit off, and,
- recalculating ratings, aside from the amount of work that implies, means everyone's rating would (probably) change, which is not something everyone will like.

dmitriking

Posts: 375
Registered: Dec 16, 2001
Age: 40
Re: provisional rating formula
Posted: Oct 13, 2017, 8:33 PM

Here's the problem I'm running into. I tried a google search but I cannot find the answer. The main issue is that I cannot find Pente.org's provisional formula anywhere on the internet. I must not be searching the right thing. I thought it was the ELO rating system, but they seem to use a different formula for provisional players.

If someone knows the name of the formula used here for provisionals, we could easily research the proper way to handle the e term that seems to be causing the problem.

watsu

Posts: 1,462
Registered: Dec 16, 2001
Home page
Re: provisional rating formula
Posted: Oct 13, 2017, 10:00 PM

My lengthy two cents - TLDR version: leave the code as it is; we've had it in place for ages and I don't see the benefit of changing it at this point.
I think the reason the provisional formula doesn't return any results (besides pente.org's FAQ page) is that Dweebo was explaining the terms and calculations in a different way than they show up in chess Elo calculations (on Wikipedia for example). Our rating system and provisional ratings system here isn't exactly the same as ones used in chess, but it is similar and I believe would fall under a general description of a modified (in some way(s)) Elo rating system. I think it's fine to give provisional players an extra boost for wins against established players; provisional is provisional and if the player sticks around to become established and plays other established players their rating will adjust fairly quickly if it is inflated.

Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat
dmitriking

Posts: 375
Registered: Dec 16, 2001
Age: 40
Re: provisional rating formula
Posted: Oct 13, 2017, 10:13 PM

For the most part, I agree - changing anything right now is probably not a good idea.

That said, Dweebo must have gotten that formula from SOMEWHERE, and the formula as written makes more sense when applied than the formula as coded.

Refer back to the original example in my first post. I was provisional 2018 and won a match against an established 1363.

The rule as written would have had me drop down to 1978.

The rule as coded increased me to 2046.

Is anyone going to argue that I should really have gained 26 points for winning a match against an opponent with a rating almost 700 points lower than mine?

The rule as written makes more sense, since it would have given me a 1978 rating.

Again though, I think changing it is a mistake, because that would be unfair to newer players, who would have a much harder time attaining a good rating after 20 games, and also, it would be adding another instance of inconsistent stat-keeping, in the same way that stats accumulated before set-based scoring are going to seem bad in comparison to stats accumulated after set-based scoring.

Either way, I think the wording should be changed to match the coding, and a note should be added to indicate that the wording has been changed but that the coding has not.

The main reason I encountered this issue is I was trying to figure out whether I would lose points even for a win, in which case I would prefer not to play the match.

rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Re: provisional rating formula
Posted: Oct 13, 2017, 10:21 PM

I changed the wording, though you make a compelling argument.

> That said, Dweebo must have gotten that formula from
> SOMEWHERE, and the formula as written makes more
> sense when applied than the formula as coded.

Not necessarily, I only know dweebo from the code he wrote and there's a ton of stuff under the hood that's a whole lot more impressive than that formula. He might have plucked it out of thin air with some solid reasoning to back it up but left us guessing as to what that was.

watsu

Posts: 1,462
Registered: Dec 16, 2001
Home page
Re: provisional rating formula
Posted: Oct 14, 2017, 1:10 AM

I did a little reading on Wikipedia about different Elo systems and the following quote leads me to suspect that Dweebo made the formula up (based on the statistical principles Elo devised, of course):
"The phrase "Elo rating" is often used to mean a player's chess rating as calculated by FIDE. However, this usage is confusing and misleading, because Elo's general ideas have been adopted by many organizations, including the USCF (before FIDE), the Internet Chess Club (ICC), Free Internet Chess Server (FICS), and the now-defunct Professional Chess Association (PCA) and Yahoo! Games. Each organization has a unique implementation, and none of them follows Elo's original suggestions precisely. It would be more accurate to refer to all of the above ratings as Elo ratings, and none of them as the Elo rating."
https://en.wikipedia.org/wiki/Elo_rating_system#Different_ratings_systems

As far as the question of arguing whether or not a 26 point gain for winning against a player with a rating 700 points lower goes, the key to remember here is that provisional ratings fluctuate a great deal at the beginning and then gradually settle down as more games are played. The next game you played against a 600 point lower player might result in losing points when you won (I have no idea whether it actually will or not, just pointing out that it might happen). So, provisional ratings gradually become more accurate as more games are played and the amount of a rating spike decreases rapidly after the first few wins.

Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat
Replies: 10   Views: 65,470   Pages: 1  
Back to Topic List
Topics: [ Previous | Next ]


Powered by Jive Software