Yace 0.20 available

[Aaron]

Geschrieben von: / Posted by: Aaron at 01 July 2000 17:33:25:
Als Antwort auf: / As an answer to: Re: EPD geschrieben von: / posted by: Dieter Buerssner at 01 July 2000 17:16:23:

I changed the tony.epd into this ..

title position 001
setboard 1r2r1k1/1pqbbppp/p2p1n2/4p3/P3PP2/2N1BB2/1PP2QPP/R4R1K b - -
Solution Rbc8
Also what are the files below generated about?
clean.ci
eclean.ci
hairy.ci
lost.ci
rh.ci
testsum.ci
I was hoping that a file would be generated that showed which move was chosen >by Yace . (not in testsum.ci??)

This should be solution. Other than the interactive commands, the test
commands are case sensitive.
These were all to optimize my private testing.
Yace has found the solution of these positions, and didn't loose it
again after more than 20000 nodes. Very easy positions (less than 10000
nodes) are not included.
Same, including also easy positions.
Solution not found or more than 10000 Nodes needed.
Solution not found
Solution not found or more than 200000 Nodes needed.
It's called test.sum and is a summary of all tests done so far, including
engine parameters, version number and more.
Just look at the log file. You may want to type
log tony.log
before doing the test.
You can also automate testing by redirecting input. Put Something like
; Default parameters
log tony.log
st 600
test tony.ci
; Try out this
pval 1.2
window 60
test tony.ci
quit
into tonytest.inp and then run
yace < tonytest.inp
or do
input tonytest.inp
at the Yace prompt.
For testing on a loaded machine, I suggest using
nodes 10M
or similar, instead of st, because this will give reproducable results.
Feel free to contribute a better description of testing for Yace
Regards,
Dieter

[Dann [not glowing] Corbit]

Geschrieben von: / Posted by: Dann [not glowing] Corbit at 01 July 2000 18:32:25:
Als Antwort auf: / As an answer to: EPD geschrieben von: / posted by: Dieter Buerssner at 01 July 2000 12:23:55:

It's part of the PGN standard by Steven J. Edwards. A web search will turn up
bazillions of copies.

I found the following:
|EPD_Spec: Extended Position Description Specification
|Revised: 1995.11.26
Is this current?
Reading it seems to indicate that support for test positions shouldn't
be difficult. Even easier might be a tool to convert EPD data into the test
format Yace/Crafty can read. Such a tool might be created with the help of Yacc (Yet another compiler compiler), the program that has inspired Yace's name,
and so might be a nice companion program for Yace.
Does such a tool already exist?
OTOH, full support for EPD will be quite a big task. Alone the "glueing code"
of Crafty is over 3500 lines.

If you don't mind C++ you could get the EpdProcessor stuff from my ftp site that was written by Thomas Mooney.
Or you could look at how any of the other amateur programs do it.
Almost every program handles EPD for input, and more than half can write EPD as output.
In addition, if you look on my ftp site, you will see dozens of epd test suites.



My FTP site

[Dann [not glowing] Corbit]

Geschrieben von: / Posted by: Dann [not glowing] Corbit at 01 July 2000 18:36:18:
Als Antwort auf: / As an answer to: Re: EPD geschrieben von: / posted by: Dieter Buerssner at 01 July 2000 17:16:23:

The positional test is different from all others. The comment c0 for each record shows the number of points that the program scores for each of the choices in the list. If the chosen move is not in the list, it scores zero points.
You add up all the points for the test suite score.


My FTP site

[Aaron]

Geschrieben von: / Posted by: Aaron at 01 July 2000 18:56:40:
Als Antwort auf: / As an answer to: Re: EPD geschrieben von: / posted by: Dann [not glowing] Corbit at 01 July 2000 18:36:18:

The positional test is different from all others. The comment c0 for each >record shows the number of points that the program scores for each of the >choices in the list. If the chosen move is not in the list, it scores zero >points.
You add up all the points for the test suite score.

I this a quick test making yace search for 10000 nodes...
And it scored 105

I'm running it again usng the recommended 10 minutes..It should score better...


Selected results (on the website http://home.interact.se/~w100107/fentest.htm)
All engines search for 10 minutes
1)Anmon5-7 AMD k6-2 450 125
2)SOS AMD k6-2 450 119
2)Rebel 10b AMDk6-2 450 119
little goliath 2000 AMD k6-2 111
Fritz6 AMD k6-2 450 107
Crafty174 k6-2 450 74