[Retros] A new program to solve fairy proof games

[CAILLAUD Michel]

Hello,
Going on testing version 0.1...
Jacobi 0.1 doesn't solve PDB P1240567
Guy Sobrecases dia5.5 r1bq1rk1/pp1sbppp/8/8/8/8/PPP1PPPP/RSBQKBSR
Hypervolage
The reason obviously is that Jacobi 0.1 ignores en passant capture in
Hypervolage (Popeye and Winchloe play it).
So, with this version, Jacobi cannot be used to test Hypervolage problems
(cooks with en passant can be missed).

On Fri, Nov 3, 2017 at 7:57 PM, François Labelle <flab at wismuth.com> wrote:

> CAILLAUD Michel wrote:
>
> As it is without rebirth, maybe you would like to incorporate it in your
>> tests :
>> dia 10.5 1sbq2sr/p1pppp1p/8/8/8/5S2/2PP2PP/bSBQKB1R condition LosingChess
>>
>
> Hi Michel,
>
> Thank you for this example in LosingChess. It shows that the slowdown
> isn't limited to Circe, something which I didn't know.
>
> Maybe in a future version, François can make "mode Tacu" as an option,
>> generating automatically the data that otherwise are subject to human error?
>>
>
> Yes, absolutely. This is why I called the trick a "workaround". My current
> plan is to automatically run every PG in "mode Tacu", so your mistake won't
> happen again. My guess is that doing so will speed up most PGs, and maybe
> slow down a small fraction of PGs that don't need it by a little.
>
> v0.1 is also doing something stupid: it's always calling both solvers even
> when the first one rejects a move. v0.2 will fix this, so the automatic
> "mode Tacu" in v0.2 should be faster than the manual "mode Tacu" in v0.1.
>
> Testing the Einstein PG on Jacobi's home page, the Tacu trick causes a 26%
> slowdown, but after the "stupidity fix" the slowdown is only 9%. This looks
> like acceptable collateral damage in order to automatically get 40x speed
> gains for other PGs.
>
>
>     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/20171104/ded0d5df/attachment.html>