Not logged inCSS-Forum
Forum CSS-Online Help Search Login
CSS-Shop Impressum Datenschutz
Up Topic Hauptforen / CSS-Forum / 42000
- - By Guenter Stertenbrink Date 2019-04-20 08:36 Upvotes 1
auf 3491

https://docs.google.com/spreadsheets/d/1XSJiCcQpCLv0fNwrUn7jXjdkZFU63YFEWpdXv6dSSg0/edit#gid=312836954

das ist schon einen thread wert (imho)
Parent - By Clemens Keck Date 2019-04-20 09:10 Upvotes 1
moin moin

die werde ich mir auch mal ansehen.
Aktuell hat die 41997 bei mir das beste Ergebnis erzielt.
https://docs.google.com/spreadsheets/d/1rS9dw2WebiCIhyMH4L7YH2ZB2b_Bi8Ac9UahuPBoQa8/edit?usp=sharing

C.K.
Parent - By Clemens Keck Date 2019-04-22 10:34
die 42000 läuft seit gestern Abend bei mir.
nach 48 Spielen liegt sie vorne !
Das bedeutet für mich immer zuerst : oh weh, ein Fehler am Testaufbau....

Code:
nn 42000, Blitz 5m+3s  2019

 
1  Lc0 v0.21.1-42000                   +29  +9/=34/-5  54.17%    26.0/48
2  Stockfish 100419 64 BMI2-x12        -29  +5/=34/-9  45.83%    22.0/48


da bin ich ja mal gespannt wie es weiter geht. Immerhin noch 52 Partien zu spielen.
C.K.
Parent - By Clemens Keck Date 2019-04-23 10:15
also zum ersten Mal gibts einen match Sieg von Lc0-42000.
Hauchdünn, ungenau sozusagen. 1 Partie "Vorsprung".
Code:
1) Lc0 v0.21.1-42000                4 :    100 (+15,=71,-14),  50.5 %

    vs.                                :  games (  +,  =,  -),   (%) :   Diff
    Stockfish 100419 64 BMI2-x12       :    100 ( 15, 71, 14),  50.5 :     +4

deshalb ein rematch auf einer anderen Maschine, sonst gleiche CPU und GPU.
aktuell nach knapp 50 Spielen liegt Stockfish eine Partie vorne, und die letzten 5 Vorgabestellungen, da holt SF bisher nochmal wichtige Punkte
Code:
nn 42000, Blitz 5m+3s  2019

 
1  Stockfish 100419 64 BMI2-x12    +8  +11/=25/-10  51.09%    23.5/46
2  Lc0 v0.21.1-42000               -8  +10/=25/-11  48.91%    22.5/46


C.K.
- By Guenter Stertenbrink Date 2019-04-27 10:35 Edited 2019-04-27 10:58 Upvotes 1
against 32930 :

mtgostark(150) , 42(5) , 29(10)
glbch(200)     , 35(5) , 25(10)
c.k.(100)      , 47(5)
sp-cc(300)     , 11(4,+19) , +26(3)
mattblach(100) , 19(5) , 23(10)
-----------------------------------------
average 2019/04/27 : +34 Elo better than 32930 in games against SF

------mtgostark(150) , 3473(5) , 3460(10)------
(32930,3431)
41550,3401
41600,3409
41650,3445
41665,3421
41700,3460
  41711,3443
41725,3419
41750,3432
41770,3448
41800,3467
41825,3461
41850,3446
41876,3449
41900,3441
41925,3430
41950,3461
  41958,3456
41975,3443
42000,3483
42025,3467
42050,3463
42075,3465
42100,3489

-----glbch(200) , 63(5) , 53(10) , 56(15)---------
(32930,28)
41770,48
41800,57
41812,71
41823,25
41830,16
41845,36
41852,43
41862,53
41871,44
41876,55
41889,86
41906,48
41917,55
41930,59
41940,58
41953,64
41965,21
41978,57
41985,44
41997,30
42021,55
42046,63
42046,63
42070,90
42100,46

------c.k.(100) , -13(5)------------
(33000,-60)
41812,-24
41889,-38
41906,-53
41965,-28
41997?,-14
42000,+03
42000,-03
42029,-24
42043,-28

----sp-cc(300) , +67(4) , 3579(3)----------------
{32930,+56,3553}
41665,+56,3577
41800,+94,3582
41978,+62,3577
41800dk,+56,3561
against average of sf,ho,ko
against 7 engines , CCRL-Elos

-----mattblach(100) , 3467(5) , 3471(10)-----------
11248,3433(50)
32930,3448
32890,3473
41631,3473
41665,3451
41680,3448
41711,3491
41800,3487
41842,3499
41876,3466
41916,3470
41958,3466
41965,3466
42017,3466
42100,

https://docs.google.com/spreadsheets/d/1XSJiCcQpCLv0fNwrUn7jXjdkZFU63YFEWpdXv6dSSg0/edit#gid=312836954
https://docs.google.com/spreadsheets/d/1681AAHzmxIIujzCXAZsPV-ycZk_BCNUKl45LquYRhSg/edit#gid=1682609908
https://www.sp-cc.de/lc0-testing.htm
https://docs.google.com/spreadsheets/d/1rS9dw2WebiCIhyMH4L7YH2ZB2b_Bi8Ac9UahuPBoQa8/edit?usp=sharing
https://docs.google.com/spreadsheets/d/1QxAG6XVTvvTAGlZ-kpSSvv0VuMGh7RkJjgKqgT37vMU/edit?ts=5bf04640#gid=1633485693

---------------------------

LuckyDay   26. Apr.
On discord there has been a lot of work done in regards to standardised testing,
with each of the nets getting tested under the same conditions for 1000-2000 games.
Atm the best performing nets were - 42100, 42016, 41997, 42010, and jhorthos'
personal T6.532 net (which is a net that he created from 42000 that was retrained
further on T40 games with 7man TB rescoring and a partial LR drop).
These 5 nets, along with 41800 as a reference point are currently having
a final round of testing prior to selection for the TCEC cup.
In the preliminary testing, 42100 did indeed perform very well.
I expect however that the jhorthos T6.532 net will be selected for TCEC cup
as it has been performing the best so far of all the nets.
If you would like to participate in this standardised testing, check out the
requirements in discord. You will need to use cutechess-cli for it and have a
computer capable of running SFdev at 4 threads. Submission deadline will
prob be in just under 2 days
--------------------------------------------------
- - By Clemens Keck Date 2019-04-27 10:52
ich hatte das match mit der 42000 wiederholt.
beide matches zusammen war die Elo +-0
Parent - By Guenter Stertenbrink Date 2019-05-04 10:14
hier das Bildchen dazu :

http://magictour.free.fr/lc0-19a.GIF

nich ganz klar, ob es noch Fortschritt gibt
Up Topic Hauptforen / CSS-Forum / 42000

Powered by mwForum 2.29.3 © 1999-2014 Markus Wichitill