Bayesian Elo rating

[Greg75]

Hi everybody.


I'm not Shure if this IS the correct place but I found no place on Internet where I could reach the father of Bayeselo.

First thing to know, I have fistfull of years in studying by myself only some méthods which are the best known : I had to manipulate matrices withthé Massey méthod , Poisson distributions had me to make my own Poisson Distribution on m'y casio graphic calculator... waiting thé next train while mine had machine troubles ...In 2 jours thé basic program began to born...Later I found that m'y graphic tableur Excel -like worked faster in sheet mode than his basic but easy high-level basic casio... Now all m'y formulas s are on sheet (liké open office or Excel) .Programs mode are slow, but I typed basic programs when I was 10 years , Always drawned nice things in graphic polar mode..

Before m'y casio graphic (even if I have Excel sheets Aldo on PC), I used DC MOTO, a Thomson french computer of the 80's 6bits processor6809 which allows more than thé MO/TO which were under-exploited.

WHY DCMOTO ? It's one of the best , if not thé best,THOMSON emulator of the Basic Microsoft which I was sofamiliar with...some Line command on Microsoft II kiné of c:/ list/ help..

And , for chose who remember thé game Numero10 , which was good animated foot game, althought thé IA of the computer was perfectible...Traditionnally little program in basic launched a waiting page before thé game itself in binary low level interprétation 'assembleur', thé more easy for a computer, thé harder to understand for an human...Hère it's not this football game itself but thé awaiting page which was in binary, but Numéro 10 had a little thing more : a .GTI image format de f Michel Platini (hé was mister n°10 ), caution for thegame (thé best on Thomson and even some Amstrad fans admitted " yeah not Bad"...

I Come back to my basic program on 6809 : I had some formulas which , I layer knew that a part was used, but now way I was on thé rightt path, and even use no more since an estimator radier and implémented with régression gives beter results...three quarters of the saison thé stats are quote enough pertinent for thé one who keeps linearrity.

For m'y basic program on MO/TO, I obsiously choose th . GTI format image of Michel PLATINI to make wait (it was no use) and for thé pleasure to have this Graet football payer while I typed datas when thé programme asked me datas.
That was m'y first coding for making oddsandtips. M'y Casio was smart in a pocket everywhere sh'd go I still use her but non more programs all on sheet excel. Estimator of nombre of goals, linear régression (m'y home own cooking) or vector/Matrix, it could even, If Ihad Time enough go to thé Skellam distribution, via Bessel they can be on thé machine.

Oh m'y apologies I write a bit too much, more for inteoducing myself than other.
I used a lot Elo (hockey sur glace, tennis ( mixed with a ratio of Victory by match played and on particuliar surfaces )and of course thé Platini's sport, football (elorating for nationals sélections, clubelo for club -only-european -)..

Elo IS simple (if tpu don't have a Big list to computer everyday) to understand and to use. Thé only trouble that it don't cares of draw match, which are 1/3 of issues.
On a Chess site (Abrobecker.free.fr ,with memoryI will give tout link layer if asked, a Chess fan) I dig a formula that , fonction of the difference of Elo coefs, gives thé probability ofa tiède , draw match, for Chess but works for football too...

I had to modify some coefficients of the formula of draw with Elo becuse if in Chess game 25% of a rien IS the maximum (and 20% thé maximum checked), in football WE need a probability of draw higher than 30%, 1/3 of gales observed IS normal.
.I Always leak m'y sources if I had no autorisation I give the name the link, anyway the coefficients for football have ssized by me according résultats of football.


And thé I Heard of Bayeselo....I knew but not feeling strong enough to use it. One dayI came back to it, downloader it, triés to understand maximisation etc...
First I downloader another thing on thé site in links, which could have helped me, and, althought I'm not a Chess fan ( play yes but watch no more than this) I downloader thé Portable Game Notation .PNG on a site which looked serions ( FIDE maybe) o fb thé game Fischer/Kasparov, one o the most inglorious.
And then , nothing Bayelso can't read...I triés to create one with few football matches and nothing more
It was last year, severals years to try it, not mypriority bit when I have thé Time I open thé case

So my questions are, I Would be glad ro h e a constructive answers:

-Can WE really simulate a football ligue with 20teams given their Elos ( m'y goal IS to compare thé beginning of thé Season to the actual current Elo)...

-Do we must typing on a .pgn all 380 matchs of the Season or today thé 300 matchs played before that day?

-WE CAN add players or team and attribuate them an Elo Coef . in thise case, iss there a model of .pgn which allows input 20 teams and their Elos, tgis Would be easie than make data of 300 matchs adding thé 80 which left...
I Would be, if this s possible, glad if a model of .pgn makes BayesElo working and computing thé end of Season of Ligue2 french (yes Hope you'll excuse m'y english I'm french born in Caen whose professional club Plays L2)...

To sum on up how IS itpossible to simulate a championship with 20 clubs (I have notes the Elo coefs at thé biginning and current day IS given by clubelo.com)

My .pgn of Fisher Kasparov game IS maybe corrupt or have Bad nomenclature I had , according to thèse notations laws triés everytging m'y download of Bayeselo don't recognize the .pgn..

Sorry for this long post, I i'troduced myself, as usual on forums for a first post, Hope somebody Would give me more précision in the case of round Robin tournament with 20 players. Chess or football it's thé same méthod..

Further one of m'y goals Would be to , starting with thé Elo coefs, giving BayesElo thé possibilité, if not yet, to take thé tièd draw matchs, giving 1 point to the teams, when thé probability of a match are P(1)=50 P(N)=30 P(2)=20...Hère if 1 Would win with40% WE have 30% la f draw IS non negligeable...and more with 30/40/30 which leads to no winner and with m'y coef of thé Abrobecker formula thé matchs which reaches 45%oftied draw have often point-shared isses...
In Chess where thé Abrobecker formula fibres 20% to 2 5% maxiIS quite really accurate but 'not adapted to soccer. I did a little change , by empirism...
Making thé Elo computing with such a formula could help BayesElo to bé more compétitive in forevasting football xhampionships. But its not m'y first step.

Any tips ? All m'y .pgn are not recognize by BayesElo.
If Only I could compute like exemples on the site I download it , I'd have make a Big step. I'm not in hurry 1 ou 2 rimes père year I try but since this Time I mist admit that I have to Ask spécialiste.

Have a nice day, or a good night.
Thank tout for attention.

[Rémi Coulom]

Hi,

Thanks for your interest in bayeselo.

Can WE really simulate a football ligue with 20teams given their Elos ( m'y goal IS to compare thé beginning of thé Season to the actual current Elo)...

bayeselo can simulate a tournament, but not with elo ratings that are input by hand. Ratings must have been computed by bayeselo from data. An example is given on the web page of bayeselo:
https://www.remi-coulom.fr/Bayesian-Elo/
See the "usage documentation" section.

-Do we must typing on a .pgn all 380 matchs of the Season or today thé 300 matchs played before that day?

If you have this data in a computer-readable format, you could write a script to convert it automatically.

-WE CAN add players or team and attribuate them an Elo Coef . in thise case, iss there a model of .pgn which allows input 20 teams and their Elos, tgis Would be easie than make data of 300 matchs adding thé 80 which left...

You cannot set elos manually in bayeselo. They have to be computed from data.

Any tips ? All m'y .pgn are not recognize by BayesElo.

Please include a sample of your pgn data, how you read it with bayeselo, and the error message you get. You can also check the PGN specs there:
https://ia902908.us.archive.org/26/item ... -03-12.txt