[Retros] A new program to solve fairy proof games

Allan Bell allanbell2 at gmail.com
Mon Nov 20 06:17:55 EST 2017


Hi  Francois,

In Demolition Mode it would be useful, especially for longer PGs, to have a
further option to specify the most likely opening moves, after which the
passes would begin.  It would artificially shorten a problem and risk
missing cooks that begin early, but such a facility would help to find
later transpositions/cooks more quickly.

Is it possible to have a "Store and Stop / Recall and Continue" facility
the way Jacobi runs?   One could then close Jacobi and resume a long test
later.  It would also be useful to back up current progress periodically to
avoid needing to start over if solving is interrupted by an unexpected shut
down (see below).

Finally, a graphic interface to input positions would be useful when you
can add it.  I use iNatch which, in the version I use, gives you the
forsyth from the diagram, but with N instead of S -- hence my desire to
have Jacobi accept N as input even though it can output S).

I have been having trouble with crashes testing a Circe PG.  I've checked
my settings, cleared cache and cookies, rebooted modem and computer before
running.  Still, there may be something about settings I've missed.  Sleep
Mode is on "never", and I don't see any other time-out setting.  Both
crashes happened after I'd left it to run unattended overnight with all
other tabs closed.  The following is some history:

   - at 8.5 Jacobi found the intended solution in Chrome (128M):
   stipulation dia8.5 forsyth
rsbqk1s1/pppppp1p/5r1p/7B/P3Rb2/4P3/1PPP1PPP/1SB1K1SR
   condition Circe
   Currently exploring (100.00%, 14277 s, 7061 nodes/sec)
   Done. Compute time = 14277.682 s
   exact solutions (1)
   1.a4 h5 2.Ra3 Rh6 3.Re3 Rf6 4.Re4 g6 5.e3 Bh6 6.Qxh5 [+bPh7] Bf4 7.Be2
   h6 8.Bd1 gxh5 9.Bxh5 [+bPh7]
   - at 10.0, using 1500M, it crashed (snap) sometime after 34000 s:
   stipulation dia10.0 forsyth
rsbqk1s1/ppppp2p/5p1p/6b1/4R1BP/4Pr2/PPPP1PP1/RSB1K1S1
   condition Circe
   - at 10.0, using default 128M, snap sometime after 19 hours and before
   23.5 hours (last noted time was 68100 s): stipulation dia10.0 forsyth
   rsbq2s1/ppppppkp/5r1p/7B/P3RP2/8/1PPP1PPP/1SB1K1SR condition Circe --
   (note change of position, but shouldn't be too significant I think --
   except maybe for solving time?

The complete problem would be 11.0, a little too long to test from the look
of it (intended last move 11.f3 Rxf4[+wPf2]).   You might note that the
idea came from your comment on doubled pawns in orthodox PGs!  The rule
about pawn capture doesn't apply in Circe of course.  I was thinking of
sending it for the RIFACE retro solving tourney next year, but not if
you're competing.  You can see it would be useful to specify 1.a4 h5 2.Ra3
Rh6 and to then start Jacobi.

Best wishes,
Allan.



On 10 November 2017 at 04:20, François Labelle <flab at wismuth.com> wrote:

> Hi everyone,
>
> Jacobi v0.2 is now available. To update to v0.2 you might have to reload
> the solver page twice (for technical reasons).
>
> This new version addresses problems and suggestions that were brought up
> on the mailing list, specifically...
>
> I wrote:
>
>> cond Hypervolage
>>
>> Jacobi v0.1 fails to see the solution 1.cxb8=S O-O.
>>
>
> This bug is fixed now.
>
> My current plan is to automatically run every PG in "mode Tacu", so your
>> mistake won't happen again.
>>
>
> This is in effect, so I hope that the solving time of PGs is now never
> significantly slower than Popeye.
>
> I also do the opposite trick: I run every Tacu PG in "mode generic" with
> similar results: some slowdowns, but at least one spectacular improvement:
> Tacu's problem on the welcome page is now solved 16x faster.
>
> Jacobi v0.1 uses a hash table with a fixed size of ~128 MB. I think I'll
>> be able to make it configurable in v0.2.
>>
>
> This is working too. You can now set the hash table size, but only once,
> before the first click on "Start". To change the hash table size again you
> have to reload the page.
>
> Experimentally the maximum is 2022M in Firefox and around 1850M in Chrome,
> but it might vary with the operating system and the browser version number.
>
> Jacobi v0.2 also fixes other problems that I discovered myself or that
> were communicated privately. For a complete list of changes, see the
> "Version history" section on the welcome page.
>
> Thank you for testing!
>
>     François
> _______________________________________________
> Retros mailing list
> Retros at janko.at
> https://pairlist1.pair.net/mailman/listinfo/retros
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://pairlist1.pair.net/pipermail/retros/attachments/20171120/2190f47d/attachment.html>


More information about the Retros mailing list