[Retros] Two little SPG-challenges

[Mario Richter]

Dear retro fans,

thanks to all who replied either here or by private mail!

Although only Noam interpreted my first challenge the way 
it was intended, it was interesting to see the solving 
times on different platforms with different programs.


Some remarks with regard to challenge 1.

It should have been read this way:


> can *you*  (by brain, not by computer) do better than ...




> Position A: rnb1kbnr/pp1ppppp/8/8/8/8/PPPP1PPP/RN2KBNR

>             SPG in 7.5 moves  (13 + 14) 


Problem A is computer-generated, so I could measure my own solving
time (approx. 4min) and make an estimate that a more experienced
solver might get the solution in about 3min.


> Position B: rnbqk1nr/pppppp2/8/8/8/8/PPPPPPP1/R1BQKB2

>             SPG in 7.0 moves  (12 + 13)


Problem B is hand-made, so - based on the assumption, that problem B 
seems to be 3 to 4 times more difficult than A, my estimate here 
was 10-15min.

So to make it a challenge I chosed the configuration "Natch on a
350MHz PII-machine" which - based on my estimates, could be beaten 
by humans.

And Noam showed that it really can be done:

Noam> I found a solution of A in 2-3 minutes,

which was what I expected,

Noam> and a solution of B in about 5.

which is far less than I expected, so my question here is:
Did you find your solution B simply by trial and error, or did you -
besides the missing Sb1 - find some clou, which helped you in
finding the solution?



Based on the solving times posted so far:

 program   hardware     hash    pos. A       pos. B      (provided by)
------------------------------------------------------------------
  natch    PII/350MHz   128M   3 mn 50s     34 mn 04s    (mri)

  Natch    P4/3 GHz     512M   1 mn 18.13s  10 mn 11.87s (Anders)

Euclide    P4/ 3,4 GHz    ?    0 mn 14 s     0 mn 24s    (Nicolas)
  Natch    P4/ 3,4 GHz    ?    4 mn                      (Nicolas)

 ?natch    AMD64 (3200+) none  0 mn 14.01s   4 mn 10.93s (Marco)

 ?natch    P4/1,4 Ghz   200M   1 mn 19s      8 mn 21s    (Thierry)

 popeye    P4/3GHz      128M  36 mn 08.875s  0 mn  7.89s (mri)

  natch    P4/3GHz      128M   0 mn 47.97s   4 mn  1.15s (mri)
                          3M   0 mn 14.34s   3 mn 22.80s (mri)

we have the following winners:

Pos A:  Euclide and Natch with approx. 14s
Pos B:  popeye with approx. 7.9s


Nicolas> "Euclide" is clearly faster than "Natch" for 
Nicolas> that kind of problems ...

What do you think?
My impression is, that Euclide and Natch are better in problems
where a 'human-style' thinking is useful for solving (Pos A), 
whereas popeye is more suitable for problems (like Pos B), where 
a 'brute-force'-approach is preferable.



Noam> ... that's not a fair comparison, because Natch is finding
Noam> a complete list of solutions and I can't claim to have proved
Noam> that there are no others.

Nor did Natch prove, that there are no shorter solutions...
(The Natch documentation explicitly states: "Only the solution in 
the given number of single moves are found.")

I would wish the programmers of Natch & Co. would provide a switch,
which forces the detection of shorter solutions.
Is there any specific reason not to do so?



For the second part of my little challenge, I will use a separate
mail.


Thanks again,

   mario