Awesome, I finally figured out how to use Bayesian elo

[swami]

I never had any kind of ratings list, can you tell me how to merge the pgn, what tools are needed and how to use them?

[Tony Thomas]

I could give you step by step directions. To merge .pgn, I used pgnscanner. Simply download pgnscanner and put it in C:/Pgnscanner. Copy all your .pgn files to that same directory. Open the pgnscanner and type the commandline "merge -dir=C:\Pgnscanner. It will merge all the pgn's in that folder and it will output a file called merge.pgn. If you get that far, let me know.

[Tony Thomas]

I am so happy that I decided to post my rating list here, I didnt edit anything, so if anything, this should be considered as an early test.

Code: Select all

Rank Name                           Elo    +    - games score oppo. draws 
   1 Naum 3                        3038   62   57   127   81%  2800   24% 
   2 Hiarcs11.2SPUCI               3014   58   54   147   75%  2807   16% 

SNIP

Jeebus, how did you manage to test so many engines? That's a mighty achievement.

Fast time controls..1m+1sec..

[Tony Thomas]

I am so happy that I decided to post my rating list here, I didnt edit anything, so if anything, this should be considered as an early test.

Great, I never figured it out
For Graham and my CM11 testing, I had to use ELOStat through Arena

Ray, I thought you were a computer wizard, being able to make a computer and all. It took me two tries to figure out how to use it. Here is the step by step directions on how to use it. Since you are able to use elostat, I am assuming that you have a single file. The hard part about all these commandline tools is that if you do not give them the exact command, they will not work properly, even a space or a comma makes the difference. I am going to call your pgn, ray.pgn..Note that I still do not know how to chose the minimum number of games.

1) open bayesian elo
2) Place ray.pgn in the same directory
3) type "readpgn ray.pgn"
4) after the console reads the pgn, type elo
5) type mm
6) type exactdist, you should be able to see it counting some numbers backwards.
7) type offset 2300 instead of 2300, you can use any starting value.
8) type ratings, it should bring up an elo list almost instantly..
9) to save the list type ratings>rays.txt, all the ratings will be saved in a file called rays.txt.
10) x
11) x
I hope this helps..

[Tony Thomas]

Is it just coincidental that you chose "Top 57"?


#57 engine could be the reason!!!

Not coincidential, my rating list should be called Rise of Romi..

The rise of ...?

Code: Select all

a = 57
b = 60
c = 63
d = 66
E = 69
f = 72
g = 75
H = 78
I = 81
j = 84
k = 87
l = 90
m = 93
N = 96
o = 99
p = 102
q = 105
R = 108
S = 111
t = 114
u = 117
v = 120
W = 123
x = 126
y = 129
z = 132

S = 111
H = 78
E = 69
R = 108
W = 123
I = 81
N = 96

111 + 78 + 69 + 108 + 123 + 81 + 96 = 666

Ha! Ha! Ha!



Just kidding people!

Time for me to make some silly goose good noodle soup.

[hgm]

I would still advise you to also try the calculation with the prior set to zero, to see how much it matters for your tournament setup. Simply repeat the procedure you describe above, but type a line

prior 0

just before giving the 'mm' command. If the ratings remain the same, you can rest assured. But if they don't, you would have to worry...

[Tony Thomas]

I would still advise you to also try the calculation with the prior set to zero, to see how much it matters for your tournament setup. Simply repeat the procedure you describe above, but type a line

prior 0

just before giving the 'mm' command. If the ratings remain the same, you can rest assured. But if they don't, you would have to worry...

I will try that when I come back from work tonight. Also, can you tell me how I can prune the list by adding a command that measures the rating of an engine with certain number of games??

[Tony Thomas]

Note that BayesElo is near-perfect for deriving ratings from round-robins between not-too-different engines, but that it leads to severely compressed rating scales for Swiss-like tournaments between engines spread out over a very wide rating range. If you have such a case (which is likely, when making a rating list) it is better to use it with prior=0 (in stead of the default prior=2).

I tried it few minutes ago as I said I would, but I am having problems. I did prior 0, then I gave the command mm, then it started iterating. After iteration 10000 4.735222e-005, it doesnt do anything. The last line is
00:00:37,61 and I dont know if it means anything. It wont let me enter exactdist or any of the rest of the commands needed to calculate a rating list and it also uses 100% of the cpu.