Alouette 0.0.1

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
Roland Chastain
Posts: 659
Joined: Sat Jun 08, 2013 10:07 am
Location: France
Full name: Roland Chastain

Re: Alouette 0.0.1

Post by Roland Chastain »

The version 1956 is buggy. It plays only illegal moves. Sorry for that. :oops:
Too late to edit my previous message and remove the link.
Qui trop embrasse mal étreint.
User avatar
Roland Chastain
Posts: 659
Joined: Sat Jun 08, 2013 10:07 am
Location: France
Full name: Roland Chastain

Re: Alouette 0.0.1

Post by Roland Chastain »

Roland Chastain wrote: Mon Apr 29, 2019 10:32 pm The version 1956 is buggy. It plays only illegal moves. Sorry for that. :oops:
Too late to edit my previous message and remove the link.
No, it works correctly. It was a problem coming from elsewhere. Not sure to have understood what happened, but it looks like Cute Chess doesn't support several engines with the same internal name. :|
Qui trop embrasse mal étreint.
tpoppins
Posts: 919
Joined: Tue Nov 24, 2015 9:11 pm
Location: upstate

Re: Alouette 0.0.1

Post by tpoppins »

tpoppins wrote: Mon Apr 29, 2019 7:05 pm Whoa, that some quick work, Roland!

100 40/2 games vs. Sapeli RND with concurrency=4 completed in 1m 8s without a single error! Bravo!
Now I just need to backport this to v0.0.3 -- since v0.0.5 scores over 90% vs. a random mover v0.0.3 should still be useful.

So far I figured out that the following needs to be inserted in interprete.pas at line 152, after { Promotion }

Code: Select all

    if (Length(ACoup) = 4) and EstUnePromotion(APos, ACoup) then
    begin
      Eteint_(LType^, LDep);
      Allume_(APos.Dames, LDep);
      LType := @APos.Dames;
    end else
but the changes in meilleur.pas are beyond me.
The above changes were sufficient to reduce stall/disconnect rate in v0.0.3 by a factor of ~10. Good enough for me.
The few test games played with the latest v0.0.5 showed no problems, certainly nothing severe like illegal moves. I'll report here if I run into any.
Good job, Roland.

I suppose you had problems when running the CLI exec of CC? The GUI won't allow separate engine entries with the same name.
Tirsa Poppins
CCRL
User avatar
Roland Chastain
Posts: 659
Joined: Sat Jun 08, 2013 10:07 am
Location: France
Full name: Roland Chastain

Re: Alouette 0.0.1

Post by Roland Chastain »

tpoppins wrote: Tue Apr 30, 2019 5:12 pm The few test games played with the latest v0.0.5 showed no problems, certainly nothing severe like illegal moves. I'll report here if I run into any.
Good job, Roland.

I suppose you had problems when running the CLI exec of CC? The GUI won't allow separate engine entries with the same name.
Thank you for your message. In fact, the problem came from a modification that I made later (after I posted the zip file). This is what happens when you should go to bed and continue to work. :)

Yes, I use cutechess-cli, but in fact the problem didn't come from Cute Chess.

Thank you for having helped me to find and solve these two bugs. :wink:
Qui trop embrasse mal étreint.
User avatar
Roland Chastain
Posts: 659
Joined: Sat Jun 08, 2013 10:07 am
Location: France
Full name: Roland Chastain

Re: Alouette 0.0.1

Post by Roland Chastain »

Hello!

Alouette 0.0.5 is available, for Linux and Windows.

An important bug has been fixed which was cause of illegal moves in some rare cases.

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   1 Moustique 0.2                 396     102      70   90.7%   18.6%
   2 Pulsar 2009 9a                147      78      70   70.0%   22.9%
   3 Alouette 0.0.5                 55      72      70   57.9%   24.3%
   4 Alouette 0.0.4                 45      74      70   56.4%   21.4%
   5 NEG 1.2                       -70      75      70   40.0%   20.0%
   6 Ram 2.0                       -76      50      70   39.3%   61.4%
   7 NEG 0.3                      -130      68      70   32.1%   35.7%
   8 Iota 0.3                     -322      81      70   13.6%   27.1%
Finished match
Qui trop embrasse mal étreint.
User avatar
Roland Chastain
Posts: 659
Joined: Sat Jun 08, 2013 10:07 am
Location: France
Full name: Roland Chastain

Alouette 0.0.6 alpha

Post by Roland Chastain »

Hello! A new version of Alouette is coming. Windows binaries are available here. If you are interested in Linux binaries, please let me know it (or else download Free Pascal). :)

The new Alouette sees four halfmoves ahead (against two halfmoves in the previous version). It recognizes the different forms of the go command and has a time management.

I removed all the little evaluation tricks which were doing nothing. The best move function lies only on the "one recursion" minimax. I just added a bonus for moving pawns first (to avoid repetitions).

I will take some days to make some more tests and polish the code.
Qui trop embrasse mal étreint.
User avatar
Roland Chastain
Posts: 659
Joined: Sat Jun 08, 2013 10:07 am
Location: France
Full name: Roland Chastain

Alouette 0.0.6

Post by Roland Chastain »

Hello! Here is Alouette 0.0.6.

Some quick and dirty results, with very short time controls (Cute Chess option "tc=40/40").

Standard variant:

Code: Select all

Rank Name                         Elo    +    - games score oppo. draws 
   1 Belofte 0.9.0                 63   89   83    40   69%  -101   13% 
   2 Alouette 0.0.6                 7   64   62    72   58%   -53    6% 
   3 Alouette 0.0.5                 0   62   60    72   57%   -50   11% 
   4 Alouette 0.0.6 recursion 0   -14   62   61    72   55%   -46    7% 
   5 Iota 0.3                    -212   74   79    40   25%   -45   50% 
   6 NEG 0.3                     -283   82   94    40   20%   -31   20% 
FRC:

Code: Select all

Rank Name                         Elo    +    - games score oppo. draws 
   1 Pulsar 2009 9a               753  287  138   108  100%     6    0% 
   2 Alouette 0.0.6                22   44   43   132   39%   222   38% 
   3 Alouette 0.0.5                 0   44   44   132   34%   230   37% 
   4 Alouette 0.0.6 recursion 0    -7   46   47   108   36%   198   42% 
   5 Luciole 0.0.3                -66   87   92    24   38%     5   42% 
If you set "recursion" to 0 in the INI file that Alouette will create, the version 0.0.6 behaves more or less like 0.0.5.

Regards.

Roland
Qui trop embrasse mal étreint.
User avatar
Roland Chastain
Posts: 659
Joined: Sat Jun 08, 2013 10:07 am
Location: France
Full name: Roland Chastain

Re: Alouette 0.0.1

Post by Roland Chastain »

Instead of an option in a INI file, I decided to make three different engines, Alouette-A, Alouette-B and... Alouette-C. :)

Maybe I will come back to the first idea, but for now I find more simple to have several executables. The idea is still to keep the different versions of the program.

Alouette-A searches at depth 1, Alouette-B at depth 2, Alouette-C at depth 4. The three versions use the same positional evaluation: bonus for pawns defending or defended by pawns, bonus for special moves, bonus for threatening opponent's king...

In my tests (tc=40/60), Alouette-B seems stronger.

Standard:

Code: Select all

360 game(s) loaded, 0 game(s) with unknown result ignored.
Rank Name               Elo    +    - games score oppo. draws
   1 Belofte 0.2.7      232   51   48   120   70%    72   15%
   2 Alouette-B 0.0.7   208   46   44   120   71%    76   38%
   3 Alouette-C 0.0.7    95   43   43   120   52%    99   38%
   4 Iota 0.3            37   39   40   120   40%   111   66%
   5 NEG 0.3             18   42   43   120   36%   114   44%
   6 Alouette-A 0.0.7     0   41   42   120   30%   118   59%
FRC:

Code: Select all

320 game(s) loaded, 0 game(s) with unknown result ignored.
Rank Name               Elo    +    - games score oppo. draws
   1 BigLion80 223x     870   84   77   120   90%   286    9%
   2 Pulsar 2009 9a     814   79   76   120   86%   299   10%
   3 Alouette-B 0.0.7   214   94   91    40   39%   407   23%
   4 Alouette-C 0.0.7   194   55   54   120   36%   444   30%
   5 Luciole 0.0.3      157   54   54   120   30%   452   29%
   6 Alouette-A 0.0.7     0   58   63   120   11%   489   23%
I try to improve Alouette-C before doing a release of version 0.0.7.

The weakness of Alouette-C has been explained here.

What do you think of my idea?
Qui trop embrasse mal étreint.
User avatar
Roland Chastain
Posts: 659
Joined: Sat Jun 08, 2013 10:07 am
Location: France
Full name: Roland Chastain

Alouette 0.0.7

Post by Roland Chastain »

Hello! Alouette 0.0.7 is available.

Alouette-C (see my previous post) has been repaired and now the hierarchy between the three engines is what I expected.

Code: Select all

Rank Name               Elo    +    - games score oppo. draws
   1 Belofte 0.2.7      217   71   67    60   67%    88   10%
   2 Alouette-C 0.0.7   201   61   59    60   69%    91   48%
   3 Alouette-B 0.0.7   173   60   58    60   63%    97   47%
   4 Iota 0.3            37   56   57    60   38%   124   62%
   5 NEG 0.3             29   58   60    60   35%   126   50%
   6 Alouette-A 0.0.7     0   58   61    60   28%   131   53%

Code: Select all

Rank Name               Elo    +    - games score oppo. draws
   1 BigLion80 223x     673  122  103    60   92%   180    7%
   2 Pulsar 2009 9a     611  111   97    60   87%   192    7%
   3 Alouette-C 0.0.7   133   70   70    60   39%   288   32%
   4 Alouette-B 0.0.7    90   68   69    60   33%   296   37%
   5 Luciole 0.0.3       64   68   69    60   29%   301   38%
   6 Alouette-A 0.0.7     0   69   74    60   20%   314   33%
tpoppins wrote: Mon Apr 29, 2019 7:05 pmNow I just need to backport this to v0.0.3 -- since v0.0.5 scores over 90% vs. a random mover v0.0.3 should still be useful.
Alouette-A is for you. :wink:
Qui trop embrasse mal étreint.
User avatar
CMCanavessi
Posts: 1142
Joined: Thu Dec 28, 2017 4:06 pm
Location: Argentina

Re: Alouette 0.0.1

Post by CMCanavessi »

Roland, I downloaded Alouette 0.0.9 and there are 2 binaries inside the zip file, alouette64 and random64. What is random64, a random mover? So kind of the same engine without eval?
Follow my tournament and some Leela gauntlets live at http://twitch.tv/ccls