Prodeo 2.2 SMP Krazy Klown

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

Adam Hair
Posts: 3226
Joined: Wed May 06, 2009 10:31 pm
Location: Fuquay-Varina, North Carolina

Re: Prodeo 2.2 SMP Krazy Klown

Post by Adam Hair »

PK wrote:Added some more comments in https://github.com/nescitus/Rodent_III/ ... params.cpp

Generally the grand plan is to move all the parameters to one big table, and then apply Texel tuning. Unfortunately, this is a lot of rather uninteresting typing, and I am rather busy nowadays, so it will take time. I already did some limited, manual experiments with tuning, and it helped with piece values.

BTW, Adam's values did not pass the test of about 3.000 games. But the games were rather interesting, so please expect "Hairy personality" in the near future.
That is because I was careless and allowed a big regression to occur:

Code: Select all

   # PLAYER                   :  RATING  ERROR   POINTS  PLAYED   (%)
   1 Rodent_III_0.195_expb    :      25      7   2680.0    5000  53.6
   2 Rodent_III_0.195_expc    :      23      8   2663.5    5000  53.3
   3 Rodent_III_0.195_expf    :       7      8   2553.0    5000  51.1
   4 Rodent_III_0.195_expe    :       7      7   2549.5    5000  51.0
   5 Rodent_III_0.195_base    :       0      3  12079.5   25000  48.3
   6 Rodent_III_0.195_expd    :      -4      7   2474.5    5000  49.5

White advantage = 15.79 +/- 2.02
Draw rate (equal opponents) = 27.26 % +/- 0.25
Rodent_III_0.195_expd represents the changes I made to the material adjustment values.
PK
Posts: 908
Joined: Mon Jan 15, 2007 11:23 am
Location: Warsza

Re: Prodeo 2.2 SMP Krazy Klown

Post by PK »

Adam, can You provide values from before the regression?
Adam Hair
Posts: 3226
Joined: Wed May 06, 2009 10:31 pm
Location: Fuquay-Varina, North Carolina

Re: Prodeo 2.2 SMP Krazy Klown

Post by Adam Hair »

The piece values are definitely a gain (represented by Rodent_III_0.195_expb) :

Code: Select all

    // Piece values 

    values[P_MID] = 86;   // was 100 
    values[N_MID] = 294;  // was 325 
    values[B_MID] = 317;  // was 335 
    values[R_MID] = 435;  // was 500 
    values[Q_MID] = 1110; 

    values[P_END] = 153;  // was 100 
    values[N_END] = 340;  // was 325 
    values[B_END] = 382;  // was 335 // decrease 
    values[R_END] = 641;  // was 500 
    values[Q_END] = 1214; // was 1000 

One thing you should know. When using CLOP and for the regression test in my previous post, I am limiting the nodes per move to 10000. I am using a 4 core laptop that I also use for internet access and so fixed nodes testing is necessary. But it is not clear if the gain I measured will hold up at typical time controls.

Also, I am using a 3 move book for CLOP and a 6 move book for the regression testing.