Alex Table Tennis - MyTableTennis.NET Homepage
  New Posts New Posts RSS Feed - Rating central
  FAQ FAQ  Forum Search   Events   Register Register  Login Login

Rating central

 Post Reply Post Reply
Author
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Topic: Rating central
    Posted: 08/14/2019 at 8:47pm
If anyone is expert on rating central rating system, can you explain the following?

at the same tourney player with 1808±32 beats player 1979±41 and gets 13 points to rating

another player with 1862±37 beats the same player 1976±41 and gets 15 points.

only 5 points difference in deviation make such the effect, while in first case point difference is bigger.?



Edited by osmar92 - 08/14/2019 at 8:48pm
Back to Top
dina_arif View Drop Down
Member
Member


Joined: 08/15/2019
Location: NY City
Status: Offline
Points: 20
Post Options Post Options   Thanks (0) Thanks(0)   Quote dina_arif Quote  Post ReplyReply Direct Link To This Post Posted: 08/15/2019 at 10:30pm
One of the basic concept that many new players forget is that the updated ratings from a tournament (or league night) are not based on just head to head results between two players only.  It is based on a given player's performance in several matches against a group of players in all events at a given tournament (or league night).

Also one of the common scenarios that happens quite often is that Player A beats B ; B beats C : C beats D  but D beats B and/or  C beats A etc.

If there were only two matches were reported as played in your example for the event you talk about, then yes it is a problem. But I doubt only 2 matches were played. Each of the 3 players in your example probably played many more matches & the net result is the final rating shown. 



Edited by dina_arif - 08/15/2019 at 11:27pm
Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 08/16/2019 at 1:54am
Originally posted by dina_arif dina_arif wrote:

One of the basic concept that many new players forget is that the updated ratings from a tournament (or league night) are not based on just head to head results between two players only.  It is based on a given player's performance in several matches against a group of players in all events at a given tournament (or league night).

Also one of the common scenarios that happens quite often is that Player A beats B ; B beats C : C beats D  but D beats B and/or  C beats A etc.

If there were only two matches were reported as played in your example for the event you talk about, then yes it is a problem. But I doubt only 2 matches were played. Each of the 3 players in your example probably played many more matches & the net result is the final rating shown. 


Ok, not many games , look what Roman gets fro beating Igor, what Gary gets for beating Igor.
Roman gets less for wider spread between him and Igor as opposed to Gary......


open image in new tab  or https://i.imgur.com/gmeVwhJ.jpg




Edited by osmar92 - 08/16/2019 at 2:03am
Back to Top
pgpg View Drop Down
Gold Member
Gold Member
Avatar

Joined: 11/18/2013
Location: United States
Status: Offline
Points: 1225
Post Options Post Options   Thanks (0) Thanks(0)   Quote pgpg Quote  Post ReplyReply Direct Link To This Post Posted: 08/16/2019 at 7:11am
IIRC, Ratings Central algorithm makes it rather difficult to isolate individual points won/lost from a given match, as opposed to USATT ELO-like approach (which is ALL about individual match impact).

Instead RC is going about that as "my prior belief about your strength was X+-Y, and given these new results I think it has to be updated to X' +- Y'". The way updating happens is NOT by considering individual matches one at a time, but rather taking them in as a whole. Very Bayesian way of thinking Wink

I suspect the question of 'how much each individual match mattered in the end' is probably not even that interesting to RC creator, but people are very used to this breakdown being available (because USATT system exists), so he came up with the way to quantify it (but probably considers it somewhat irrelevant :) )

Some more detailed explanations are available here:


USATT: ~1870
Donic Defplay - AK47 blue - Dtecs OX
Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 08/16/2019 at 10:31am
Originally posted by pgpg pgpg wrote:

IIRC, Ratings Central algorithm makes it rather difficult to isolate individual points won/lost from a given match, as opposed to USATT ELO-like approach (which is ALL about individual match impact).

Instead RC is going about that as "my prior belief about your strength was X+-Y, and given these new results I think it has to be updated to X' +- Y'". The way updating happens is NOT by considering individual matches one at a time, but rather taking them in as a whole. Very Bayesian way of thinking Wink

I suspect the question of 'how much each individual match mattered in the end' is probably not even that interesting to RC creator, but people are very used to this breakdown being available (because USATT system exists), so he came up with the way to quantify it (but probably considers it somewhat irrelevant :) )

Some more detailed explanations are available here:



I understanding the logic of algorithm, it make sense in many aspects, however. I cannot explain why in my example player with wider rating spread deserves less points having only 3 points in deviation difference with the other.

And in Not "Playing as a Way to Gain Points" i do not agree with the logic, It implies that frequency of playing beats skipping the tourney.
Firstof, the more you play regardless of rating you can lose to a weaker player or to have a bad streak.
second, System should not not perceive that if you dont attend tourney the confidence about your level of playing changes, For example, Once can continue practicing and maintain his level.
I saw many times when players dont play for a year or more and play one tourney and jump by 100 points or so for beating players they are not far in strength whom they lost and won before. 

Another example, I have low deviation 32 and play inconsistent so I stay pretty much at the same spot. I will skip for 2 years practicing and playing at different clubs , come back play well and jump much more than if I continue playing inconsistent. 


Also, in his example numbers look wrong.

From what I see in my league deviation increases by around 50 points in a year

2017-11-21DTTC November League 2017 (III)1894±94+581952±83
2016-10-04DTTC October League 2016 (I)1888±44−21886±40

here 5 month, looks at jump in deviation
2014-12-30DTTC December League 2014 (V)1896±68−21894±57
2014-07-01DTTC July League 2014 (I)1884±43+81892±39

another player example, 1 year 57 points increase

2014-05-20DTTC May League 2014 (III)1909±89+952004±84
2013-05-28DTTC May League 2013 (III)1896±46+61902±42

look how from 2013 to 2015 deviation grew from 46 to 121 and so many points gained in winning. 
SO lets say positive streak kicked in and I gained a lot. Now I could stop just for sake of staying high.

2015-08-04DTTC August League 2015 (I)2013±121+662079±106
2014-05-20DTTC May League 2014 (III)1909±89+952004±84


Edited by osmar92 - 08/16/2019 at 10:50am
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/01/2019 at 7:49pm
> If anyone is expert on rating central rating system

I know a little about it.

> at the same tourney player with 1808±32 beats player 1979±41 and gets 13
> points to rating
>
> another player with 1862±37 beats the same player 1976±41 and gets 15
> points.

Here is an offline calculation assuming that the laws are normal:

Initial Ratings
Player 1:   N(1808, 32^2)
Opponent 1: N(1979, 41^2)
Player 2:   N(1862, 37^2)
Opponent 2: N(1976, 41^2)

Player 1 beats Opponent 1: New rating for Player 1 is 1821 with a st. dev.
of 32. Gain of 13 points; more precisely 13.2171112357503.

Player 2 beats Opponent 2: New rating for Player 2 is 1877 with a st. dev.
of 36. Gain of 15 points; more precisely 15.1852970849732.

So, it agrees with the numbers Roman posted. You can check the calculation
yourself, with some effort. My math intuition is pretty good, but I always
check it with a calculation.

Now, let's try it with Player 1 having a st. dev. of 37:

Initial Ratings
Player 1:   N(1808, 37^2)
Opponent 1: N(1979, 41^2)
Player 2:   N(1862, 37^2)
Opponent 2: N(1976, 41^2)

Player 1 beats Opponent 1. New rating for Player 1 is 1825 with a st. dev.
of 37. Gain of 17 points; more precisely 17.4489615899142.

Player 2 beats Opponent 2: New rating for Player 2 is 1877 with a st. dev.
of 36. Gain of 15; more precisely 15.1852970849732.

> One of the basic concept that many new players forget is that the updated
> ratings from a tournament (or league night) are not based on just head to
> head results between two players only. It is based on a given player's
> performance in several matches against a group of players in all events
> at a given tournament (or league night).

The other matches that your opponent plays will be taken into account, but
this happens before your match with the opponent is processed. In other
words, the rating of your opponent that is shown next to your result
already includes this. That's why the opponent's rating may be different
from the opponent's rating at the beginning of the event.

> Also one of the common scenarios that happens quite often is that Player
> A beats B ; B beats C : C beats D but D beats B and/or C beats A etc.

Perhaps. I don't know how common it is. But, I'm not sure what the
relevance is to the current question.

> If there were only two matches were reported as played in your example
> for the event you talk about, then yes it is a problem.

I don't see the problem.

> IIRC, Ratings Central algorithm makes it rather difficult to isolate
> individual points won/lost from a given match

The website displays the points won/lost from each match (multiple matches
with one component are combined and the points distributed among the
matches). Of course, if the matches were processed in a different order,
the numbers might change, but the final rating wouldn't. That's just the
way belief works, whether done by a person or by a computer.

> The way updating happens is NOT by considering individual matches one at
> a time, but rather taking them in as a whole. Very Bayesian way of
> thinking

It does them one at a time. (Actually, it does each opponent one at a time.
This only matters if you play the same opponent more than once in the
event, and the result isn't that different from what you would get doing
even those one at a time. There is an example on the website.)

The only way I know to do them all at once would be to do something like
use a numerical algorithm to find the posterior mode. I don't think this
would be practical for a large rating system, but it would be interesting
to try on a small data set for comparison (there are some comparisons like
that in my paper).

> It implies that frequency of playing beats skipping the tourney.

I'm not sure what that means, but I don't think the webpage says that.

> second, System should not not perceive that if you dont attend tourney
> the confidence about your level of playing changes,

That's rather an odd thing to say. It certainly isn't the way people think.
If I haven't seen a player in a year, I will be less sure of their
level than I was a year ago.

If you are sure that a player's level is not changing, then you would
weight matches they played long ago the same as matches they play today. Of
course, we don't do that when evaluating players. More recent data should
count more.

> Another example, I have low deviation 32 and play inconsistent so I stay
> pretty much at the same spot. I will skip for 2 years practicing and
> playing at different clubs , come back play well and jump much more than
> if I continue playing inconsistent.

If you want to get a high rating without actually getting better, you
should stop playing, since every time you play, we find out you haven't
really gotten better. So, you should not play for a few years, then come
back and play only one match against a much better player. If you get lucky
and win, we will think that you are really good (but we will realize it is
only one match, so you may have gotten lucky). Unfortunately, you will have
to retire because if you keep playing, we will see that you haven't really
gotten better.

This is all just common sense, and works just as well on people as it does
on computers.

> Also, in his example numbers look wrong.

Which number is wrong? Did you calculate or just eyeball it?

> From what I see in my league deviation increases by around 50 points in a
> year

As it says on the website, the variance increases by 79.4^2 a year. This is
basic probability theory: The new law is the old law plus the random walk
plus the Poisson jump process. Since they are independent, the variances
add.

Your first example has standard deviations of

   2016-10-04: 40
   2017-11-21: 94

This is 413 days. So the second standard deviation should be

   sqrt( 40^2 + 79.4^2 * 413 / 365).

Excel says this is 93.45280371. So, that checks (the first standard
deviation is probably slightly more than 40).
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 9:28am
I wrote:

> You can check the calculation yourself, with some effort.

If you don't want to code it yourself, other people have posted
calculators:

http://usv-indigo.at/ranglisten/punkterechner/

https://www.sites.google.com/site/portmacquariett/table-tennis-ladder

> My math intuition is pretty good, but I always check it with a
> calculation.

If you did want to try to understand it qualitatively, I suggest you start
by graphing the laws.

> (there are some comparisons like that in my paper).

Correction: there are some similar comparisons in my paper, but not exactly
this type of comparison.

Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 10:47am
Thanks for the reply. Couple of questions

<Here is an offline calculation assuming that the laws are normal:>

is law normal? can we assume otherwise?

<That's rather an odd thing to say. It certainly isn't the way people think.
If I haven't seen a player in a year, I will be less sure of their
level than I was a year ago.>
so this looks subjective, imho, can we not adjust deviation on length of not playing but let the system decide what deviation to assign upon the first tourney results  after the break in playing. It seems like contemporaneous valuation of level.

Which number is wrong? Did you calculate or just eyeball it?

i just looked at it in linear way. 

I have no problem with system calculations, my issue is concept and intuitively sometimes it feels odd that small deviation of 3 points and  lesser spread give more adding in rating to player who beat the same opponent. 
Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 11:02am
<It implies that frequency of playing beats skipping the tourney.>

Therefore, there is no advantage to not playing. In other words, you cannot game the system by not playing. Just get out there and play, and the ratings will take care of themselves.

Not exactly , agree. it does not say per se. But from my real experience it differs.

Woudnt you agree that after certain age level cannot be substantially increased(basically not changed) only worsen if a player stopped playing? also stopping playing and not attending are two different categories, so the law can be skewed.
Once again, i have no doubt in your mathematical modeling I just have reservation in assumed concept.


Edited by osmar92 - 09/02/2019 at 11:06am
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 11:19am
> is law normal?

Which law? Initial laws are normal by construction. After that, they will
never be exactly normal, but they usually look reasonably bell-shaped. The
jump process introduces an asymmetry, which you can see if you graph the
laws (although it isn't as large as you might expect; I was a bit surprised
the first time I graphed it).

> can we assume otherwise?

I don't understand the question.

> so this looks subjective, imho, can we not adjust deviation on length of
> not playing but let the system decide what deviation to assign upon the
> first tourney results after the break in playing. It seems like
> contemporaneous valuation of level.

I'm sorry, but I don't understand what you wrote. Perhaps you are mixing up
what happens to the player (how their playing strength changes with time)
with what data we have on them (their match results). What happens to the
player is independent of what data we have. This is true whether it is a
person evaluating the player or a computer.

Except for the jump process, the structure of the model is really very
simple. People have used the same model (with different parameters) in
other sports, for example in chess and online gaming. I believe I'm the
first to add a jump process. The tournament surgery part of the algorithm
is also unique.

> i just looked at it in linear way.

Many things are not linear.

> it feels odd that small deviation of 3 points

Have you graphed the laws and drawn your own picture of what the updated
laws should be?

Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 11:25am
Originally posted by osmar92 osmar92 wrote:


Woudnt you agree that after certain age level cannot be substantially increased(basically not changed) only worsen if a player stopped playing? also stopping playing and not attending are two different categories, so the law can be skewed.
Once again, i have no doubt in your mathematical modeling I just have reservation in assumed concept.

I'm sorry, but I don't understand what you wrote or what relevance it has
to the discussion on the webpage https://www.ratingscentral.com/PointChangeExamples.php
Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 11:41am
> so this looks subjective, imho, can we not adjust deviation on length of
> not playing but let the system decide what deviation to assign upon the
> first tourney results after the break in playing. It seems like
> contemporaneous valuation of level.

I'm sorry, but I don't understand what you wrote


What I mean is ---- assuming system does not have less confidence when player has not participated for a period of time, so deviation is not changed. But the first time player plays after the break and based on results it calculates whether the deviation should be changed.
For example.
If a player after the year of not participating show the same results he had prior the break in playing his deviation stays the same. only if  he beats players with a level he could not beat before or lose to a  player with level he had not lose before system can assume that his level changed and he deserves to have higher deviation.

the difference between what you assume and I ponder is.
You--- system is not sure of the level hence it corrects deviation

mine- system already knows the level based on history and consider nothing is changed untill after the break a player shows his results. If nothing is changed then why would hу deserve to have deviation adjusted during a period of not playing.

<Have you graphed the laws and drawn your own picture of what the updated
laws should be?>

i havent, i will do it some day.


<Which law? Initial laws are normal by construction.>
I meant that it is not normal in our real life, skewed always. But I will take your assumption.


Edited by osmar92 - 09/02/2019 at 11:43am
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 11:57am
Originally posted by osmar92 osmar92 wrote:

What I mean is ---- assuming system does not have less confidence when player has not participated for a period of time, so deviation is not changed. But the first time player plays after the break and based on results it calculates whether the deviation should be changed.
For example.
If a player after the year of not participating show the same results he had prior the break in playing his deviation stays the same. only if  he beats players with a level he could not beat before or lose to a  player with level he had not lose before system can assume that his level changed and he deserves to have higher deviation.

the difference between what you assume and I ponder is.
You--- system is not sure of the level hence it corrects deviation

mine- system already knows the level based on history and consider nothing is changed untill after the break a player shows his results. If nothing is changed then why would hу deserve to have deviation adjusted during a period of not playing.
It appears that you think that the standard deviation is a property of the
player. It isn't. The thing that is part of the player is the playing
strength. It is certainly possible that the player's playing strength
hasn't changed, but it makes no sense to say that, if we haven't seen the
player for say a year, we are just as sure of the player's playing strength
now as we were a year ago. No person would say that. (The law and mean are
also not properties of the player.)

Originally posted by osmar92 osmar92 wrote:

<Which law? Initial laws are normal by construction.>
I meant that it is not normal in our real life, skewed always. But I will take your assumption.

By "skewed" do you mean what the word means in probability? I don't know of
any examples where it would be useful to use a skewed distribution for a
player's initial law. Can you show us a player whose initial law should be
skewed?
Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 12:08pm
Originally posted by David Marcus David Marcus wrote:

Originally posted by osmar92 osmar92 wrote:

What I mean is ---- assuming system does not have less confidence when player has not participated for a period of time, so deviation is not changed. But the first time player plays after the break and based on results it calculates whether the deviation should be changed.
For example.
If a player after the year of not participating show the same results he had prior the break in playing his deviation stays the same. only if  he beats players with a level he could not beat before or lose to a  player with level he had not lose before system can assume that his level changed and he deserves to have higher deviation.

the difference between what you assume and I ponder is.
You--- system is not sure of the level hence it corrects deviation

mine- system already knows the level based on history and consider nothing is changed untill after the break a player shows his results. If nothing is changed then why would hу deserve to have deviation adjusted during a period of not playing.
It appears that you think that the standard deviation is a property of the
player. It isn't. The thing that is part of the player is the playing
strength. It is certainly possible that the player's playing strength
hasn't changed, but it makes no sense to say that, if we haven't seen the
player for say a year, we are just as sure of the player's playing strength
now as we were a year ago. No person would say that. (The law and mean are
also not properties of the player.)

Originally posted by osmar92 osmar92 wrote:

<Which law? Initial laws are normal by construction.>
I meant that it is not normal in our real life, skewed always. But I will take your assumption.

By "skewed" do you mean what the word means in probability? I don't know of
any examples where it would be useful to use a skewed distribution for a
player's initial law. Can you show us a player whose initial law should be
skewed?

 <It is certainly possible that the player's playing strength
hasn't changed, but it makes no sense to say that, if we haven't seen the
player for say a year, we are just as sure of the player's playing strength
now as we were a year ago.>

It is subjectivity that his level changed if we havnt seen him. You havent seen him does not mean he did not train or played somewhere else.
Objectivity is only when he plays first after the break.

<Can you show us a player whose initial law should be
skewed?>

I am, playing inconsistently, can beat much higher level player and then play as usual. I can deliver surprise in results more often than others.
Veterans and youngsters will have different laws I think as opposed to peak years players
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 12:30pm
> > It is certainly possible that the player's playing strength hasn't
> > changed, but it makes no sense to say that, if we haven't seen the
> > player for say a year, we are just as sure of the player's playing
> > strength now as we were a year ago.
>
> It is subjectivity that his level changed if we havnt seen him. You
> havent seen him does not mean he did not train or played somewhere else.
> Objectivity is only when he plays first after the break.

I said that we are less sure of a player's playing strength if we haven't
seen them in a year.

You then said that it is possible that a player's playing strength has
changed, e.g., because they trained or played somewhere else.

So, you are agreeing that the longer we haven't seen a player, the less
sure we are of that player's playing strength. (And, this is exactly the
way the system thinks.)

> > Can you show us a player whose initial law should be
> > skewed?

> I am, playing inconsistently, can beat much higher level player and then
> play as usual. I can deliver surprise in results more often than others.
> Veterans and youngsters will have different laws I think as opposed to
> peak years players

I'm sorry, but this only seems to be saying that different players
should have different initial laws. It doesn't say why a skewed law would
be useful.

There is nothing in the system that attempts to model how consistent a
player is. If you beat all the 1500 players and lose to all the 1900
players you will end up with the same rating as if you lose to all the 1500
players and beat all the 1900 players. This is a simplification of reality,
but models are always simplifications of reality. Despite this (or because
of this), models can be useful.

Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 12:43pm
So, you are agreeing that the longer we haven't seen a player, the less
sure we are of that player's playing strength. (And, this is exactly the
way the system thinks.)

Yes we are less sure as humans( however, I might think that nothing changed if a player maintain his level somewhere else. I know players who do not play often in tourneys but keep training so they level stays the same)  but why dont we judge it after the fact not , lets not assume anything.
I understand your view , you want the system to adjust itself by first widening the deviation .
I ponder what will happen if system does not assume but only adjust after the first playing after the break when it can assess the reality.

I wonder what will happen if you dont change deviation with time of not playing but adjust rating somehow similar to how  it is done by USTTA or any other way  , everything else stays the same.
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 12:51pm
Originally posted by osmar92 osmar92 wrote:

I ponder what will happen if system does not assume but only adjust after the first playing after the break when it can assess the reality.

The system does not assume. It does exactly what you wrote.

As I already said, you are confusing what is real with what we know. As
long as you continue to mix these up, you will continue to be confused.

Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 12:56pm
Ok, i might confuse things.
But will you be able to run simulation of the system without adjusting deviation based on time of not participating.?
just wonder if the results in ranking players will be different.
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 1:01pm
Originally posted by osmar92 osmar92 wrote:

But will you be able to run simulation of the system without adjusting deviation based on time of not participating.?
just wonder if the results in ranking players will be different.

Of course they will be different. That's a huge change. Matches played ten
years ago would have equal weight with matches played last month. Imagine
what this would do to the rating of a player who was only a child ten years
ago, and is now playing 1500 points better.

It is odd that you would suggest such a change. One of the problems that
led to development of the system is that the USATT system has no memory and
so considers ratings from ten years ago to be just as reliable as ratings
from yesterday. Because the USATT system has no memory, they are forced to
let the system be too reactive so that players who are actually getting
better will have their ratings go up fast enough.



Edited by David Marcus - 09/02/2019 at 1:06pm
Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 1:13pm
Definitely 10 years ago for a chilв would make a big diff.
Lets do then short period , like dynamic window of memory, 1 year
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 1:49pm
Originally posted by osmar92 osmar92 wrote:

Definitely 10 years ago for a chilв would make a big diff.
Lets do then short period , like dynamic window of memory, 1 year

That's pointless. The current model increases it continuously.

Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 2:06pm
increasing would refer to more history so it would definitely smooth out fluctuation but would lessen the weight of current status. Am I wrong?
Back to Top
David Marcus View Drop Down
Super Member
Super Member
Avatar

Joined: 10/28/2010
Location: Somerville, MA
Status: Offline
Points: 121
Post Options Post Options   Thanks (0) Thanks(0)   Quote David Marcus Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 2:11pm
Originally posted by osmar92 osmar92 wrote:

increasing would refer to more history so it would definitely smooth out fluctuation but would lessen the weight of current status. Am I wrong?

By "increasing", I meant increasing the variance of the player's law with time. I suggest you read and understand


Then you will be able to answer your own questions.

Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 09/02/2019 at 2:23pm
ok, thx for now
Back to Top
osmar92 View Drop Down
Super Member
Super Member


Joined: 02/14/2012
Location: USA
Status: Offline
Points: 365
Post Options Post Options   Thanks (0) Thanks(0)   Quote osmar92 Quote  Post ReplyReply Direct Link To This Post Posted: 11/07/2019 at 1:21pm
I have something to ask you about again
Back to Top
 Post Reply Post Reply
  Share Topic   

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 12.01
Copyright ©2001-2018 Web Wiz Ltd.

This page was generated in 0.156 seconds.

Become a Fan on Facebook Follow us on Twitter Web Wiz News
About MyTableTennis.NET | Forum Help | Disclaimer

MyTableTennis.NET is the trading name of Alex Table Tennis Ltd.

Copyright ©2003-2019 Alex Table Tennis Ltd. All rights reserved.