Alex Table Tennis - MyTableTennis.NET Homepage
  New Posts New Posts RSS Feed - USATT ratings calculation
  FAQ FAQ  Forum Search   Events   Register Register  Login Login

tabletennis11.com
Forum Home Forum Home > General > General

USATT ratings calculation

 Post Reply Post Reply
Author
notgood View Drop Down
Super Member
Super Member


Joined: 04/19/2006
Status: Offline
Points: 139
Post Options Post Options   Thanks (0) Thanks(0)   Quote notgood Quote  Post ReplyReply Direct Link To This Post Topic: USATT ratings calculation
    Posted: 03/08/2018 at 3:36pm
So someone was asking me about calculating ratings, and I figured it would just be faster to write a program to do it.  However, the logic I found for the rating process gives me these three choices:

1) you gain < 51 points (just add the points to your rating)
2) you gain 51 - 75 points (add the points to your rating and then process it again)
3) you gain > 75 points and had at least 1 loss (use formula they provide)

So I'm unsure of what happens when you gain > 75 points without a single loss.  If anyone knows the answer that would be great.  If not, I was hoping to find an example on the USATT site of someone who has accomplished this feat so I can try and compare results.  Can someone point me to their result on the USATT site if they have done this at some point? 
Viscaria
Vega Pro
Hammond Pro Beta
Back to Top
mts388 View Drop Down
Gold Member
Gold Member


Joined: 03/21/2014
Location: Sonora CA
Status: Offline
Points: 1682
Post Options Post Options   Thanks (0) Thanks(0)   Quote mts388 Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 4:19pm
Look up Maria Tran who Played in the Oklahoma City tournament earlier in the year.

She went in with a 1134 rating and went 12-0.  Her wins included 1741, 1689 & 1649 players.  She was adjusted to 1399 and ended up with a 1641 rating.  I don't understand how her rating would end up below 3 of the players she beat.
Back to Top
notgood View Drop Down
Super Member
Super Member


Joined: 04/19/2006
Status: Offline
Points: 139
Post Options Post Options   Thanks (0) Thanks(0)   Quote notgood Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 4:57pm
I'm having a tough time figuring out how this works as well.  Only one of the players she played was adjusted (1401 to 1479).  So that should simplify things.

method 1:
If you just add up the points she got from wins it is 388.  1134 + 388 = 1522.

method 2:
I would adjust her to 1522 and then run the ratings again, which would put her at 1657. 

method 3:
And the > 75 point gain formula doesn't work because she doesn't have a loss:
Pass_2_rating = ( BestWin + WorstLoss ) / 4 + ( Pass_1_result ) / 2

So definitely thank you for the example since I was struggling to find one.  However, I'm still confused as to how to process this case.
Viscaria
Vega Pro
Hammond Pro Beta
Back to Top
skip3119 View Drop Down
Premier Member
Premier Member


Joined: 02/24/2006
Location: somewhere
Status: Offline
Points: 6960
Post Options Post Options   Thanks (0) Thanks(0)   Quote skip3119 Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 6:16pm
Originally posted by mts388 mts388 wrote:

Look up Maria Tran who Played in the Oklahoma City tournament earlier in the year.

She went in with a 1134 rating and went 12-0.  Her wins included 1741, 1689 & 1649 players.  She was adjusted to 1399 and ended up with a 1641 rating.  I don't understand how her rating would end up below 3 of the players she beat.
==================
In pre- and post-tournament:
  
3 players rated higher than her in pre-tournament.
Only 1 player rated higher than her in post-tournament.
Back to Top
emihet View Drop Down
Platinum Member
Platinum Member


Joined: 09/22/2009
Location: CALIFORNIA
Status: Offline
Points: 2140
Post Options Post Options   Thanks (0) Thanks(0)   Quote emihet Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 7:18pm
Originally posted by skip3119 skip3119 wrote:

Originally posted by mts388 mts388 wrote:

Look up Maria Tran who Played in the Oklahoma City tournament earlier in the year.

She went in with a 1134 rating and went 12-0.  Her wins included 1741, 1689 & 1649 players.  She was adjusted to 1399 and ended up with a 1641 rating.  I don't understand how her rating would end up below 3 of the players she beat.
==================
In pre- and post-tournament:
  
3 players rated higher than her in pre-tournament.
Only 1 player rated higher than her in post-tournament.
yes both pre and post ratings have to be looked at...in this case i don't know, but also all the other matches that were played in that tournament obviously


Edited by emihet - 03/08/2018 at 7:18pm
Stiga Clipper Wood, Viscaria,BBC,
Back to Top
pgpg View Drop Down
Silver Member
Silver Member
Avatar

Joined: 11/18/2013
Location: United States
Status: Online
Points: 906
Post Options Post Options   Thanks (0) Thanks(0)   Quote pgpg Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 7:45pm
For better or worse, the USATT rating adjustment procedure is what it is - a pretty complex beast that tries to handle a bunch of edge cases as good as it can, but does not succeed 100% of the time. 

Therefore, it is relatively easy to find examples where it does not do what you and other reasonably minded people expect it to do - e.g. example quoted above. However: trying to focus on the edge cases and fix the issue that happens 0.1% of the time is not a good strategy, me thinks. Especially since a couple of extra tournaments will fix the issue anyway for the player in question.

To the OP - I suspect you already know about this page:

https://www.teamusa.org/usa-table-tennis/ratings/rating-system

My advice is not to try to recreate the algo as described on the page linked above, since I'm not sure what exactly would be the benefit of that. For a 'quick and dirty' estimate simple table of points gained/lost is more than enough, and if you suspect you are going to be adjusted - just wait until results are processed. You probably should not be focused on ratings too much in the first place Wink

TLDR: writing the program won't be faster since you'll spend 95% of time (and much longer than you estimated in the first place) covering edge cases according to poorly specified algorithm that also happens to depend on the ratings and records of other players (multipass, fun...). Just like any other software project... Geek

USATT: ~1800
Dr. N Barricade - AK47 blue - Air UpUpUp (0.6mm)
Back to Top
pongfugrasshopper View Drop Down
Silver Member
Silver Member
Avatar

Joined: 03/22/2015
Location: USA
Status: Offline
Points: 672
Post Options Post Options   Thanks (0) Thanks(0)   Quote pongfugrasshopper Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 7:51pm
Originally posted by emihet emihet wrote:

Originally posted by skip3119 skip3119 wrote:

Originally posted by mts388 mts388 wrote:

Look up Maria Tran who Played in the Oklahoma City tournament earlier in the year.

She went in with a 1134 rating and went 12-0.  Her wins included 1741, 1689 & 1649 players.  She was adjusted to 1399 and ended up with a 1641 rating.  I don't understand how her rating would end up below 3 of the players she beat.
==================
In pre- and post-tournament:
  
3 players rated higher than her in pre-tournament.
Only 1 player rated higher than her in post-tournament.
yes both pre and post ratings have to be looked at...in this case i don't know, but also all the other matches that were played in that tournament obviously

I think the idea is that just because you beat someone higher rated than you doesn't necessarily make you higher rated than they are.  This is true both in tournament and club play.  You have to prove it with more tournament play.
Back to Top
pongfugrasshopper View Drop Down
Silver Member
Silver Member
Avatar

Joined: 03/22/2015
Location: USA
Status: Offline
Points: 672
Post Options Post Options   Thanks (0) Thanks(0)   Quote pongfugrasshopper Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 7:54pm
Originally posted by notgood notgood wrote:

So someone was asking me about calculating ratings, and I figured it would just be faster to write a program to do it.  However, the logic I found for the rating process gives me these three choices:

1) you gain < 51 points (just add the points to your rating)
2) you gain 51 - 75 points (add the points to your rating and then process it again)
3) you gain > 75 points and had at least 1 loss (use formula they provide)

So I'm unsure of what happens when you gain > 75 points without a single loss.  If anyone knows the answer that would be great.  If not, I was hoping to find an example on the USATT site of someone who has accomplished this feat so I can try and compare results.  Can someone point me to their result on the USATT site if they have done this at some point? 

This falls under the all wins scenario.  In this case, you take the median of the opponents PRE-tournament rating to get the adjusted rating.  From the adjusted rating, use the rating table to get the post tournament rating.

EDIT: Take the aforementioned tournament results for player, Maria Tran.  Her opponents PRE-tournament ratings are as follows:

783 935 995 1063 1279 1398 1401 1516 1596 1649 1689 1741

Since it's an even number of values, take the average of the middle values:
average of 1398 and 1401 is 1399 (as an integer).


Edited by pongfugrasshopper - 03/08/2018 at 8:00pm
Back to Top
LUCKYLOOP View Drop Down
Platinum Member
Platinum Member
Avatar

Joined: 03/27/2013
Location: Pongville USA
Status: Offline
Points: 2822
Post Options Post Options   Thanks (0) Thanks(0)   Quote LUCKYLOOP Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 8:55pm
Originally posted by mts388 mts388 wrote:

Look up Maria Tran who Played in the Oklahoma City tournament earlier in the year.

She went in with a 1134 rating and went 12-0.  Her wins included 1741, 1689 & 1649 players.  She was adjusted to 1399 and ended up with a 1641 rating.  I don't understand how her rating would end up below 3 of the players she beat.


There is a lot of ratings deflation in the current system. She should have been rerated to 1743 to help stop the deflation. No one should have lost rating points to her either.
Hntr Fl / 4H & BH Xiom Sigma Pro 2 2.0
Yinhe T-2 / 4H Xiom Sig Pro 2 2.0 BH Xiom Omega IV Elite Max
Gam DC / 4H DHS Hurricane 8 39deg 2.1 BH GD CC LP OX
HARDBAT / Hock 3 ply / Frenshp Dr Evil OX
Back to Top
notgood View Drop Down
Super Member
Super Member


Joined: 04/19/2006
Status: Offline
Points: 139
Post Options Post Options   Thanks (0) Thanks(0)   Quote notgood Quote  Post ReplyReply Direct Link To This Post Posted: 03/08/2018 at 11:42pm
Thanks all!  I actually was not using the page that was linked by pgpg, and on that page it has my answer:

If a player has either, all wins, or all losses, the Adjusted Rating is derived by taking the median implied rating for all of the player’s games. The implied rating is calculated using each of the opponents’ Pre-Tournament Ratings, and the Rating Chart above.

I was using a random website (http://members.tripod.com/netta_ct/ratings/ratingsprocess.html) that had copied the old version of the ratings process, and it did not have this line on the page. 


Viscaria
Vega Pro
Hammond Pro Beta
Back to Top
Gordy View Drop Down
Silver Member
Silver Member


Joined: 12/04/2011
Location: United States
Status: Offline
Points: 574
Post Options Post Options   Thanks (0) Thanks(0)   Quote Gordy Quote  Post ReplyReply Direct Link To This Post Posted: 03/09/2018 at 6:08pm
One other important thing...

In the case of all wins where the implied median rating is less than the pre-tournament rating, the adjusted rating will be set at the pre-tournament rating plus the PASS1 ratings point gain. This can only happen if someone plays so many matches against players relatively close to them that they can gain more than 75 points against players that don't push their implied median rating above their pre-tournament rating. If it sounds complicated, sorry. What we are trying to do is avoid the situation where someone can actually have a higher adjusted rating by losing a match rather than all wins.
JOOLA Nobilis / SP (FH)/ Rhyzm Tech (BH)
Back to Top
 Post Reply Post Reply
  Share Topic   

Forum Jump Forum Permissions View Drop Down



This page was generated in 0.141 seconds.
Mark all posts as read :: Delete cookies set by this forum

Cookies and JavaScript must be enabled on your web browser in order to use this forum


Copyright © 2003-2013 MyTableTennis.NET - All Rights Reserved. Disclaimer