|
|
| (1) Posted by Marcin Sterkowiec [Thursday, Sep 12, 2024 18:38] |
New software for solving chess moremovers
I would like to invite you to familiarize yourself with a program dedicated to solving chess moremovers. You can download a free version at jgisland.pl. It has an extensive test suite that can be viewed at https://jgisland.pl/download/reports/report_978g_vs_972o_byAuthor.html (it is a small database of chess composition). The most recent developments (compositions addressed) are enumerated in readme file: https://jgisland.pl/download/readme_EN.txt . The main goals of the program are speed, robustness and simplicity. I do hope you will like it.
|
|
| (2) Posted by Marcin Sterkowiec [Saturday, Sep 14, 2024 15:22] |
BTW: Here is the link to the pdf document that describes the main features of this program: https://jgisland.pl/download/JGIslandBrochure.pdf
|
|
| (3) Posted by Thomas Maeder [Sunday, Sep 15, 2024 18:31] |
Hi Marcin
Thanks - this all sounds very interesting!
Your program seems most useful to composers who currently are using Gustav. Have you made comparisons between the two programs?
|
|
| (4) Posted by Marcin Sterkowiec [Sunday, Sep 15, 2024 21:33] |
Hi Thomas,
I didn't make such direct comparisons since I am not a user of Gustav but judging by some entries in chess problem database pdb.dieschwalbe.de that specifiy time of computer analysis by Gustav, the speed of J.G.Island is at least very competitive. Also C++ seems slightly more suitable for time-critical tasks than Pascal.
To make a quick evaluation by yourself you can also take a look at the following view https://jgisland.pl/download/reports/moremovers_979_vs_978g_byTotalTime.html that starts from the most demanding problems (sorted by analysis time in milliseconds descending). There are really not many problems that are solved by J.G.Island in more than 30 seconds (only about 3.5% of all the moremovers included in the test suite). Of course it would take much more time, if run in "all solutions" mode (to detect a possible dual).
If information on https://www.chessprogramming.org/Gustav is up to date (about capacity of Gustav to solve up to #163 K7/p3p1p1/1p2p1p1/rp4B1/brpn4/Rnk5/8/4N1N1 by Ottó T. Bláthy), then J.G.Island seems a small step further because it can solve #210 rBb5/P2p4/P2Ppp1p/1B1p1p1B/2prn3/KRnk4/3p4/b4N1N (by the same great Hungarian engineer and composer). At the same time, I have to admit that many times the work by Olaf Jenkner let me notice weak points in J.G.Island and inspired to make appropriate enhancements; a trace of this can be seen for example in comments here: https://pdb.dieschwalbe.de/search.jsp?expression=PROBID=%27P1410779%27 . I do believe that J.G.Island can be a good competitor and that this competition will be advantageous both for our software and for chess composition.
|
|
| (5) Posted by Olaf Jenkner [Monday, Sep 16, 2024 01:21] |
Hi Marcin,
I have a Gustav version written in C and compiled by gcc. It has no GUI and is therefore not easy to handle.
Standard Gustav is written in Free Pascal (Lazarus) and runs about 20-30% slower than C-Gustav.
In August 2023 (Gustav 4.2 e) the maximal length of problems was extended to 210 moves.
Gustav solves the #210 in about 20 minutes with parameters "flies=0, black moves=17". The C version
was running 15:30,44 min on my laptop. It took me some time to find out the best parameter 17.
Did your program find the parameters on it's own? 1143312 msec is nearly the same time.
|
|
| (6) Posted by Marcin Sterkowiec [Monday, Sep 16, 2024 11:32] |
Hi Olaf,
Yes, J.G.Island has an initial phase (analysis of position type) in which it tries to find an optimal set of all parameters for the run (if needed, it can be overridden manually). This initial phase is single threaded and usually is very fast (less than 100 msec), but I saw a few special cases where it takes a few seconds - usually when program suspects Sea Snake position type after some intro. However I still have pending cases to be covered (e.g. #87 http://www.yacpdb.org/?id=278008 with quite an original intro 1. Qd2 B:b4 2. Qc3 Ra5 - so in my test suite for now there is only #85 K7/p5p1/1pN3p1/rp2p3/kbp5/qrQ5/4N1p1/n2B2Rb). Anyway, sea snake itself has a very specific way of handling, settings are completely different, actually even paths of code are completely different than in other position types.
|
|
| (7) Posted by Hauke Reddmann [Monday, Sep 16, 2024 17:41] |
I would extend the suggestion to compare:
Compare *all* problem solving engines (might include databases, might include
a standard LiChess Stockfish engine) for various aspects. There is the classic
Alybadix, there is the completely unknown MateMaster (which I still use
for composing #2, it's a bit easier than Popeye/Olive) etc...
|
|
| (8) Posted by Marcin Sterkowiec [Tuesday, Sep 17, 2024 11:54] |
I have addressed your suggestion in a separate thread: https://matplus.net/start.php?px=1726566653&app=forum&act=posts&fid=gen&tid=3195&pid=26209#n26209 (Creation of benchmark(s) for software solving orthodox moremovers)
For me it seems quite a challenge actually, worth a separate thread...
|
|
No more posts |
MatPlus.Net  Forum General New software for solving chess moremovers |
ISC 2024
