STS re-re-re-re-re-visited

Posts : 14 Join date : 2023-01-10

A new installation did not solve the problem. The solution was a bit tricky:

1) The temere.zip file that can be downloaded from your webside does not contain a file named make-html.bat

2) Adding this file in the temere folder did not solve the problem.

3) I also had to add the instruction "make-html.bat" at the end of the run-temere.bat file. After doing this everything worked better (but not the same as your output)

STS re-re-re-re-re-visited - Page 2 Clipbo14

Subject: Re: STS re-re-re-re-re-visited Wed Jan 11, 2023 8:12 am

Operation goes via the batch-file STS.BAT

Code:: set MT=100
set HASH=64
set THREADS=1
set PROTOCOL=uci
set EPD=epd\test.epd
set MRL=mea_results

set EXE=engines\Koivisto_8.0.exe
set NAME=Koivisto_8.0
mea.exe --engine %EXE% --name %NAME% --hash %HASH% --threads %THREADS% --protocol %PROTOCOL% --epd %EPD% --movetime %MT%
mrl %MRL% %MRL% %EPD% %MT%

set EPD=epd\test.epd

Change the location of your epd

And double click STS.BAT

Posts : 13 Join date : 2022-10-23

peter wrote:: How about that:
These 594 "best" (not too easy) positions out of 1500 of STS

https://www.dropbox.com/s/x4qmr9ucykk0r76/594.epd?dl=0

together with about 500 UHO (or of even less balanced Anti Draw)- opening postions out of Stefan Pohl's opening- sets?

https://www.sp-cc.de/uho_2022.htm

http://talkchess.com/forum3/viewtopic.php?p=941241#p941241

Posts : 14 Join date : 2023-01-10

First I tried the following:

1. Loaded down the file from your homepage and unzipped it
2. run the bat files as it comes with no changes nowhere
3. Result

STS re-re-re-re-re-visited - Page 2 Clipbo15

so strage result out of the box without touching anything.
So my conclusion is
a) the zip file from the homepage is broken or
b) my PC has something special that causes that behaviour.

I also tried the sts.bat as you sent. Result is better, but used time and solving time is not correct.

STS re-re-re-re-re-visited - Page 2 Clipbo16

If possible, and you can test steps 1 to 3 and get another result, than it can only be on my PC.

Posts : 14 Join date : 2023-01-10

At last I could make it work propperly. So I can say, the Zip file that can be downloades from

https://rebel13.nl/misc/mea.html

is broken.

The MEA tool that comes from one of this two links

http://rebel13.nl/dl/sts-lc0.7z

http://rebel13.nl/dl/sts-sf15.7z

is working propperly.

Posts : 13 Join date : 2022-10-23

peter wrote:: together with about 500 UHO (or of even less balanced Anti Draw)- opening postions out of Stefan Pohl's opening- sets?

https://www.sp-cc.de/uho_2022.htm

Here I collected 548 out of Stefan Pohl's
UHO_2022_6mvs_+180_+189.pgn
and
UHO_2022_8mvs_+180_+189.pgn
about equally mixed positions after 5, 6 and 7 moves.
To be downloaded in set together with sts-SF15.epd as 2048- position suite in MEA- syntax from here

https://www.dropbox.com/s/i24egoj6wz3f719/uhost.epd?dl=0

Results so far again seem to be even a little bit nearer to VSTC game playing ones then MEA-STS already is:

Code::   EPD : uhosts
   Time : ms
      Max   Total Time Hash
   Engine Score Found Pos ELO Score Rate ms Mb Cpu
1 SF230302 24509 1628 2048 4149 26584 92.2% 100 8 8
2 SF15 24429 1634 2048 4135 26584 91.9% 100 8 8
3 lc0230217-325M 24065 1598 2048 4072 26584 90.5% 200 2 2
4 Dragon3.2 23984 1586 2048 4059 26584 90.2% 100 8 8
5 Berserk11.1      22949 1501 2048 3883 26584 86.3% 100 8 8
6 Rebel16.2    22099 1424 2048 3739 26584 83.1% 100 8 8

   Created with MEA
   by
   Ferdinand
   Mosca

I have a bigger subset of opening positions and another one of 256 endgame positions only too, and I sometimes let them run all together in one suite of 2345 positions all in all, but I prefer having three singular runs for opening, midgame and endgame instead, with different TCs for different sets and different engine- pools.
That gives more exact answers to more exact questions. Wanting to get exact answers to exact questions as for game playing results one has to have different kinds of game playing matches at all as well, also with different hardware-TCs, different pools of opponents and different opening positions. Something like "overall playing strength" is elosion only anyhow, was always and is so even more nowadays
Smile

Posts : 13 Join date : 2022-10-23

peter wrote:: I have a bigger subset of opening positions and another one of 256 endgame positions only too, and I sometimes let them run all together in one suite of 2345 positions all in all

https://www.dropbox.com/s/z4og7ozuochu6ox/2345.epd?dl=0

e.g.:

Code::    EPD : epd\2345.epd
   Time : 100ms
      Max   Total   Time Hash
   Engine Score Found Pos ELO Score Rate ms Mb Cpu
1 SF230203 29374 1898 2345 4072 32444 90.5% 100 8 8
2 SF15 29096 1879 2345 4036 32444 89.7% 100 8 8
3 Dragon3.2    28518 1819 2345 3955 32444 87.9% 100 8 8
4 lc0230217-325M   28237 1814 2345 3915 32444 87.0% 200 2 2
5 Berserk11.1 27004 1719 2345 3744 32444 83.2% 100 8 8
6 Ethereal14 26349 1665 2345 3654 32444 81.2% 100 8 8
7 Rebel16.2 26133 1633 2345 3622 32444 80.5% 100 8 8
8 Kovisto9.2 25842 1656 2345 3586 32444 79.7% 100 8 8

   Created with MEA
   by
   Ferdinand
   Mosca

Subject: Re: STS re-re-re-re-re-visited Mon Feb 27, 2023 1:04 am

It's almost for real.

Posts : 13 Join date : 2022-10-23

Better as for determinism of course is single thread (3.5GHz), thus A-B-engines can be compared directly (with same TC) to LC0 at two threads of CPU and 3070ti Nvidia GPU, value at "Hash" there stands for NN-cache instead.
And then 200msec. instead of 100 give again even a little bit nearer to VSTC game playing comparable results:

Code::   EPD  : epd\2345.epd
   Time : 200ms
     Max    Total Time Hash
   Engine Score Found Pos ELO Score Rate ms Mb Cpu
1 SF230203 28537 1838 2345 3960 32444 88.0% 200 4 1
2 SF15 28500 1826 2345 3951 32444 87.8% 200 4 1
3 lc0230217-325M 28237 1814 2345 3915 32444 87.0% 200 2 2
4 Dragon3.2      27679 1763 2345 3838 32444 85.3% 200 4 1
5 Berserk11.1    27391 1757 2345 3798 32444 84.4% 200 4 1
6 Ethereal14 26920 1723 2345 3735 32444 83.0% 200 4 1
7 Rebel16.2     26847 1689 2345 3721 32444 82.7% 200 4 1
8 Koivisto9.2 26702 1716 2345 3703 32444 82.3% 200 4 1
9 Wasp6.50 26070 1632 2345 3618 32444 80.4% 200 4 1
10 PowerFritz18 25484 1615 2345 3532 32444 78.5% 200 4 1

   Created with MEA
   by
   Ferdinand
   Mosca