D***e
发帖数: 48486 | 1 To our USATT Members:
Over the past several months, we have received several comments regarding a
perceived gap in our ratings system "adjustment algorithms" whereby a player
could actually gain ratings points by losing an individual match.
I, along with several other members of the USATT community, have spent the
last several weeks studying the issue, and come to the conclusion that this
"gap" needs to be fixed.
As a result, earlier today I authorized the implementation of an change to
our adjustment algorithms which address the key "gap" in the calculation
procedure. In short, whereas the previous algorithms did not adjust a player
whose final adjusted rating resulted in a lower rating than their pre-
tournament rating, the change in the algorithm that was made today will now
adjust that player to an adjusted rating based on their "points based"
ratings gain.
This change will not affect players whose net ratings gain (based on net
points gained) is between 51 and 74 points -- only those whose gain is
greater than 75 points.
While I realize this sounds a bit confusing, it is more easily explained by
example:
EXAMPLE 1
?Player X's pre-tournament rating is 1949. Player X goes 15-0, with a "
points based" ratings gain of 77 points. The median implied rating of Player
X's wins is 1940.
?Old Algorithm: Under the old algorithm, Player X would not receive a final
adjustment, as the median implied rating of the wins was lower than their
pre-tournament rating.
?New Algorithm: Under the new algorithm, Player X will receive a final
adjusted rating of 2026. That is based on the "points based" adjustment,
which is derived by adding their pre-tournament rating of 1949 and their
points gain of 77. Thus, the final adjusted rating is 1949 + 77 = 2026.
EXAMPLE 2
?Player X's pre-tournament rating is 1949. Player X goes 10-3, with a "
points based" ratings gain of 77 points. Using the old algorithm (average of
best win/worst loss and pre-tournament rating), the adjusted rating would
result in a rating of 1940.
?Old Algorithm: Under the old algorithm, Player X would not receive a final
adjustment, as the adjusted rating (after taking the average of best win/
worst loss and pre-tournament rating) would be lower than Player X's pre-
tournament rating.
?New Algorithm: Under the new algorithm, Player X will receive a final
adjusted rating of 2026. That is based on the "points based" adjustment,
which is derived by adding their pre-tournament rating of 1949 and their
points gain of 77. Thus, the final adjusted rating is 1949 + 77 = 2026.
The new algorithms were deployed at 2pm (MT) today, and will affect
tournaments going back to January 15, 2015 (when the switch was made from
NATT to RailStation).
On a related note, we received word this afternoon that there is a minor
issue with the calculations which is currently resulting in some ratings
being incorrectly adjusted downward. As our ratings system and algorithms
are not intended to allow for downward adjustments, we are proactively
working to address this issue and expect to have the problem resolved soon.
We understand that the ratings deadline for the US Open is approaching on
Friday, and expect to have the issue resolved before then. In the event that
the error has not been fixed by that time, we will adjust the ratings
deadline accordingly so that our members playing in the US Open are not
adversely or negatively affected.
I would encourage you to reach out to me ([email protected]
/* */) if you have
any questions about the change to the algorithm -- or to Tiffany Oldland (
[email protected]
/* */) with any other ratings related questions.
Respectfully,
Gordon Kaye
CEO
USA Table Tennis | s**e
发帖数: 1834 | 2 也收到了这个usatt的email，很高兴他们开始做些算法调整。
但这个email有两点问题。
1. 这次算法调整的中心是为了纠正老算法有些情况下有”输了球可以（多）涨分“的
现象，但他的两个例子只是说新老算法调分区别，可没有说老算法是怎么就”输了球
可以（多）涨分“了。
2. 这个可能只是笔误。他说涨了75+后，老算法调分用”taking the average of best
win/worst loss and pre-tournament rating"。而实际上老算法调分用的是”taking
the average of best win/worst loss and pass1 rating"。 | D***e
发帖数: 48486 | 3 其实我看了看，但是竟然没看懂
呵呵
best
taking
【在 s**e 的大作中提到】
: 也收到了这个usatt的email，很高兴他们开始做些算法调整。
: 但这个email有两点问题。
: 1. 这次算法调整的中心是为了纠正老算法有些情况下有”输了球可以（多）涨分“的
: 现象，但他的两个例子只是说新老算法调分区别，可没有说老算法是怎么就”输了球
: 可以（多）涨分“了。
: 2. 这个可能只是笔误。他说涨了75+后，老算法调分用”taking the average of best
: win/worst loss and pre-tournament rating"。而实际上老算法调分用的是”taking
: the average of best win/worst loss and pass1 rating"。
|
|
