What is New in Toga II 1.3x4?

[IWB]

maybe you don't understand.
it happens in ANY tournament i do.

what you test if you run matches is mainly learning and book influence.

1. Do you really think that I am testing with learning switched on?

If so it seems you either underestimate a lot of people or you are a bit too confident about yourself!

2. Obviously you like to post something without reading what the others have written. I do not quote myself but please read this about my matchconditions: http://216.25.93.108/forum/viewtopic.ph ... 61&t=12896

Whats left of learning and book influence? Nothing!

Greetings
Ingo

[mclane]

do you think that the ELO LIST guys switch learning off ?

what is the value of 100 diced elo list games ?

(did you get the impression that i talked about you?)

If so it seems you either underestimate a lot of people or you are a bit too confident about yourself!

is this sentence meant as a question or as an insult ?

2. Obviously you like to post something without reading what the others have written. I do not quote myself but please read this about my matchconditions: http://216.25.93.108/forum/viewtopic.ph ... 61&t=12896

i do NOT read it ingo.
i was speaking in general.

if the elo-list guys switch learning OFF (that would imply they
change the DEFAULT settings of the programs), they would
get doubles and tripples and and and.
than they would have to throw those games out.
this would imply they do book-keeping .

Whats left of learning and book influence? Nothing!

in the rating lists there is much left.
in a tournament where the opponents change each game,
the learning/book-line-learning mechanism cannot be as succesful.

and here we have the reason tournaments get different results
than matches eng-eng.

you say toga1.2.1a is stronger than 1.3x4. i do say the opposite.

if i would have had ANY tournament saying the opposite,
i would post it.

if it would be random, 1.2.1 would be stronger in one tournament, and 1.3 in the other. but this does not happen.

another example

Code: Select all

    Motor                      Punkte RyToHiPxFrLoSpPyGlthCoToNoPr    S-B
01: Rybka v2.3.lk.x64          8,5/9  ·     1 1 1 1 =   1   1 1 1   30,25
02: TogaII1.3x4 [default]      7,5/9    ·   = = 1 1 1 1     1 = 1   26,00
03: Hiarcs11.1UCI              7,5/9      · =     0 1 1 1 1 1 1 1   19,00
04: program x                  6,0/9  0 = = · = =     1   1   1 1   20,00
05: Fruit-061115a              6,0/9  0 =   = ·     1 = 1 =   1 1   18,75
06: LoopMP 12.32               5,0/9  0 0   =   ·   = = 1 1 = 1     13,50
07: Spike1.2                   4,5/8  0 0 1       · = = = 1 1       16,25
08: program y                  4,5/9  = 0 0   0 = = ·   1 1 1       15,50
09: Glaurung121-EM64T          3,0/8    0 0 0 = = =   ·       = 1    9,25
10: theBaron1.8.1Uci           3,0/9  0   0   0 0 = 0   · 1 =   1    6,00
11: Colossus2006f              2,0/9      0 0 = 0 0 0   0 · =   1    4,75
12: TogaII1.2.1a               1,5/8  0 0 0     = 0 0   = = ·        5,00
13: Now0704                    1,0/8  0 = 0 0 0 0     =       · 0    5,25
14: ProDeo1.2                  1,0/9  0 0 0 0 0       0 0 0   1 ·    1,00

61 Partien von 182 gespielt

Intel6300, 25'+10", ponder:off

the only difference between these tournaments is, that the later
has the faster hardware, and the slower time controls.
and the gap between the results is even bigger.
this could indicate that with more computation time the gap increases.

[Uri Blass]

<snipped>

i was speaking in general.

if the elo-list guys switch learning OFF (that would imply they
change the DEFAULT settings of the programs), they would
get doubles and tripples and and and.
than they would have to throw those games out.
this would imply they do book-keeping .

No

It does not work like that.

The elo guys from CEGT and CCRL play without book from predefined position(I understood that ingo also use similiar conditions).

They practically do something like the nunn match between every pair of programs except the fact that they use different random positions so you cannot expect a lot of doubles.

Uri

[Graham Banks]

if the elo-list guys switch learning OFF (that would imply they
change the DEFAULT settings of the programs), they would
get doubles and tripples and and and.
than they would have to throw those games out.
this would imply they do book-keeping

Duplicate games are discarded by CCRL. Can't speak for others.

[mclane]

not ssdf

[mclane]

?
if you play from predefined positions, how should duplicate games happen ?

[Graham Banks]

?
if you play from predefined positions, how should duplicate games happen ?

We use generic opening books with a 12 move limitation rather than opening suites.
Here are our 40/40 testing conditions.
http://kd.lab.nig.ac.jp/chess/discussio ... ced43e514f

[Uri Blass]

?
if you play from predefined positions, how should duplicate games happen ?

We use generic opening books with a 12 move limitation rather than opening suites.
Here are our 40/40 testing conditions.
http://kd.lab.nig.ac.jp/chess/discussio ... ced43e514f

It is practically similiar to what I suggested:

I wrote the following:
"They practically do something like the nunn match between every pair of programs except the fact that they use different random positions so you cannot expect a lot of doubles."

having generic opening book with a 12 move limitation means that
the position out of book is random position and the only difference between it and what I suggested is that the programs do not have to play the same random position both with white and black.

Uri

[mclane]

data update

Code: Select all

    Motor                      Punkte RyToHiPxFrSpLoPyGlTothCoNoPr    S-B
01: Rybka v2.3.lk.x64          10,0/11 ·   = 1 1 1 1 =   1 1 1 1 1   44,25
02: TogaII1.3x4 [default]      9,0/11    ·   = = 1 1 1 1 1 1 = = 1   38,50
03: Hiarcs11.1UCI              9,0/11  =   · = 1 0   1 1 1 1 1 1 1   35,50
04: Program x                  7,0/10  0 = = · = 1 =   1     1 1 1   31,00
05: Fruit-061115a              6,5/11  0 = 0 = ·   = 1 =   1 = 1 1   25,25
06: Spike1.2                   6,5/11  0 0 1 0   ·   = = 1 = 1 1 1   23,00
07: LoopMP 12.32               6,5/11  0 0   = =   · = = = 1 1 1 1   20,75
08: Program y                  5,5/10  = 0 0   0 = = · 1 1 1 1       24,50
09: Glaurung121-EM64T          4,5/11    0 0 0 = = = 0 · = 1   = 1   15,75
10: TogaII1.2.1a               3,0/10  0 0 0     0 = 0 = · = = 1      9,25
11: theBaron1.8.1Uci           3,0/11  0 0 0   0 = 0 0 0 = · 1   1    8,25
12: Colossus2006f              2,5/11  0 = 0 0 = 0 0 0   = 0 ·   1   10,25
13: Now0704                    1,0/10  0 = 0 0 0 0 0   = 0     · 0    6,75
14: ProDeo1.2                  1,0/11  0 0 0 0 0 0 0   0   0 0 1 ·    1,00

75 Partien von 182 gespielt

1.3x4 as weak as Toga1.2.1a ?

[Terry McCracken]

data update

Code: Select all

    Motor                      Punkte RyToHiPxFrSpLoPyGlTothCoNoPr    S-B
01: Rybka v2.3.lk.x64          10,0/11 ·   = 1 1 1 1 =   1 1 1 1 1   44,25
02: TogaII1.3x4 [default]      9,0/11    ·   = = 1 1 1 1 1 1 = = 1   38,50
03: Hiarcs11.1UCI              9,0/11  =   · = 1 0   1 1 1 1 1 1 1   35,50
04: Program x                  7,0/10  0 = = · = 1 =   1     1 1 1   31,00
05: Fruit-061115a              6,5/11  0 = 0 = ·   = 1 =   1 = 1 1   25,25
06: Spike1.2                   6,5/11  0 0 1 0   ·   = = 1 = 1 1 1   23,00
07: LoopMP 12.32               6,5/11  0 0   = =   · = = = 1 1 1 1   20,75
08: Program y                  5,5/10  = 0 0   0 = = · 1 1 1 1       24,50
09: Glaurung121-EM64T          4,5/11    0 0 0 = = = 0 · = 1   = 1   15,75
10: TogaII1.2.1a               3,0/10  0 0 0     0 = 0 = · = = 1      9,25
11: theBaron1.8.1Uci           3,0/11  0 0 0   0 = 0 0 0 = · 1   1    8,25
12: Colossus2006f              2,5/11  0 = 0 0 = 0 0 0   = 0 ·   1   10,25
13: Now0704                    1,0/10  0 = 0 0 0 0 0   = 0     · 0    6,75
14: ProDeo1.2                  1,0/11  0 0 0 0 0 0 0   0   0 0 1 ·    1,00

75 Partien von 182 gespielt

1.3x4 as weak as Toga1.2.1a ?

I didn't know Toga II 1.2.1a was weak