Comments/Ratings for a Single Item

Even if I manually plot the good position, the tournament handler, makes it wrong and then it starts- wrongly.

H.G.

I'm not sure how things are at your end but I've just seen a time loss, but now it's rare it was the first in rougly 30 games, It's playing at 2mins/30moves.

Any ideeas to help me set up automatic tournaments? Should I send you somehow what I have for you to check?

Ok, the time thing happend twice in 22 games, BAdddd! I think my computer is stupider because it's slower, so what exactly should I do?

Once again the time lost 3/24, That's really bad, Of course I can just ignore those results as I still do things manually!

I guess I'll switch to 5mins+5secs, it never lost on time with added time as time trouble has a slightly different meaning here.

No that doesn't work either, I never seen time trouble in 15mins+15secs in 20 games now (not even with the old program), but that's simply to slow!

I think, I managed to solve the time trouble, It camed from cloning Fmax by copying to a  second exe file.I can't explain it why but it consumed a lot of resources, by comparison to the fairy max.exe proper. I have to play more games but it seems fine, now.

Nope, that was not it, still getting in time trouble at game 4!

This should work, if you have the Fairy-Max folder as sub-folder inside the folder where winboard.exe is.

In winboard.ini:

/settingsFile=settings.ini
/saveSettingsFile=settings.ini
/fcp=fmax.exe
/fd=Fairy-Max
/scp=fmax.exe
/sd=Fairy-Max
/variant=elven
/testLegality=false
/pieceToCharTable="PNBRQW.A..C......GKpnbrqw.a..c......gk"
/loadPositionFile="./start.fen"

And in start.fen (in the WinBoard folder):

r2wccw2r/1anbqkbna1/pppppppppp/10/10/10/10/PPPPPPPPPP/1GNBQKBNG1/R2WCCW2R w - - 0 1
r2wccw2r/1gnbqkbng1/pppppppppp/10/10/10/10/PPPPPPPPPP/1ANBQKBNA1/R2WCCW2R w - - 0 1

And in the fmax.ini file (in the Fairy-Max folder):

// Large-board variant
Game: elven
10x10=3
7 3 4 5 10 11 9 4 3 7
7 3 4 6 10 12 8 4 3 7
p:100 -16,24 -16,6 -15,5 -17,5
p:100 16,24 16,6 15,5 17,5
n:340 14,7 31,7 33,7 18,7 -14,7 -31,7 -33,7 -18,7 46,6 50,6 -46,6 -50,6 35,6 29,6 -29,6 -35,6
b:350 15,3 17,3 -15,3 -17,3
w:440 47,7 49,7 -47,7 -49,7 19,7 13,7 -13,7 -19,7 15,7 -15,7 17,7 -17,7
c:450 1,7 16,7 30,7 34,7 -1,7 -16,7 -30,7 -34,7 2,7 32,7 -2,7 -32,7
R:470 1,3 16,3 -1,3 -16,3
A:780 16,3,17 16,3,15 -16,3,-15 -16,3,-17 1,3,17 -1,3,15 -1,3,-17 1,3,-15
G:830 17,3,16 15,3,16 -15,3,-16 -17,3,-16 17,3,1 -15,3,1 -17,3,-1 15,3,-1
Q:950 1,3 16,3 15,3 17,3 -1,3 -16,3 -15,3 -17,3
k:-2  1,7 16,7 15,7 17,7 -1,7 -16,7 -15,7 -17,7
k:-2  1,7 16,7 15,7 17,7 -1,7 -16,7 -15,7 -17,7
#
# P& fmWfceFifmnD
# N& NmZ
# W& CF
# C& WAD
# G& FyafsF
# A& WyafsW
# K& K

You then only have to tick "step through lines orpositions", and specify a "Rewind after" of 2 in the Tournament Options dialog, and select "Machine Match" in the Mode menu. (After selecting the desired Time Control.)

 

It Works! The problem was that everything was in 1 folder!

So, it works, now I'm going to setup pawn odds games in order to see how strong the pawns are and then games aancas vs griffins, aancas vs queens, griffins vs queens and equivalent for apothecary 2.

I think archbihop vs queen will lead to expected results.

I'm so excited , thanks for the opportunity!

H.G.

FYI: In your apothecary 2 variant pawn promotes to zebra.

H.G.

If I have the time defeat disabled, is there a way to find out if negative time was ever an issue in my experiments?

H.G.,

And one more question:

Is there a way to see partial results?

I meant otherwiselly than writing a small c++ program that prelucrates the final string from the *.trn file

WinBoard displays the current score of a match in the title bar, behind the engine names. (You might not see it if the engine names are too long, and you use a small board size.)

If Auto-Flag is off, there is no way to see if time went negative. With Fairy-Max 5.0b3 this should not happen often enough to degrade the results significantly, however.

Btw, I just got the following idea: due the due to the large number of pieces in Apothecary compared to FIDE games last pretty long (and time usage per move varies more wildly). One way around this would be to start from positions that do not have all 14 pieces, but just a subset of 8, and pick those such that the total set of positions contains the pieces in the correct ratios. E.g. to make 11 start positions for testing Griffins vs Aancas you take 5 Apothecary armies plus 6 extra Kings, Griffins and Aancas and 60 Pawns. Each position gets a King, Griffin, Aanca and 10 Pawns. The remaining pieces you divide over the 11 positions sort of randomly, with the restriction that there should never be more than 2 of any kind, and never more than one Queen. This would give shorter games, but, more importantly, I would expect that it also magnifies the effect of any imbalance, e.g. deletion of a Pawn, so that you do not need so many games topush the statistical noise down to the desired fraction of a Pawn.

Well, I did not see any time trouble lately so I think we are ok in 5.02b3. I'll think I'll stick to all pieces (-odds pawn) for now because of that!

H.G.

Kevin Pacey's link makes me wonder. How do we take into account the engine strength in our experiments? Suppose we make n experiments at 40moves/1min and n experiments at 40moves/2mins.We get obviously a better pawn in the second experiment. How does this relate to the real strength of the pawn, whatever that means?

About the games being tedious, Ive just seen a 280 moves apothecary 2 game KRBPvsKA endgame. Quite cool!

The method is not very sensitive for the level of play (Elo of the engine,or thinking time per move). Sure, a stonger player has a higher success rate in converting a material advantage into a win. But that is not only true  for the advantage of a Pawn, but also for other material advantages, such as the exchange or having two Griffins instead of two Aancas. When you devide them through each other to convert the advantage back to Pawn units, this cancels out. Provided you stay in the range where scores are still linear, and do not saturate (say 25%-75%). Pawn-odds in 8x8 Chess at the level of Fairy-Max 40 moves/2 min gives about 68%, so well within the range. But you can always measure material imbalances that are matches to better than half a Pawn.

H.G., My pawn odds tests are almost completed, but I think that I am going to redo them as I think I found a way to implement the promo rule în fairy-max. Here how it goes : I use a second piece, say pawn 2, to optionally promote the pawn to at rank 8, then because pawn2 can GO only further at rank 9 he Will promote there to The apropiate piece or a pawn 3. Is that possible? If so how will The AI react to IT? Which value should I use for The Power of pawn2 and pawn3? I'm proposing bishop value for pawn2 and slightly below rook value for pawn3.

I am not sure what exactly you propose, but I am pretty sure it won't work. Fairy-Max only promotes piece #1 (in the white promotion zone) and #2 (in the black zone). And it always assumes promotion is mandatory and changes piece #1 into piece #7, and piece #2 into piece #7 or #9 (the latter only if black does not start with a piece #7). Except in Grant Acedrex, which is sort of a hard-coded exception, where the Pawns promote to the piece that started on their promotion square. So different Pawns can promote differently there, but there still never is any choice to what a particular Pawn promotes to.

It is this no-choice limitation that is the show stopper. It is really deeply engrained in the way Fairy-Max works: moves are internally encoded by a from-square and a to-square, and no other info.

Ok! So back to square 1!

I have finished the pawn odds experiments for both apothecary 1 and apothecary 2 games.

The experiments were setup in the following way:

In the bishops inside initial setup a8 pawn gets deleted

In the bishops inside initial setup a3 pawn gets deleted

In the knights inside initial setup a8 pawn gets deleted

In the knights inside initial setup a3 pawn gets deleted

In the bishops inside initial setup b8 pawn gets deleted

In the bishops inside initial setup b3 pawn gets deleted

In the knights inside initial setup b8 pawn gets deleted

In the knights inside initial setup b3 pawn gets deleted

repeat for each column until j for a total of 40 games

repeat for 25 times.

Total number of games=1000

 

Apothecary 1 results:

normal setup side wins: 529

draws:124

deleted pawn side wins:347

normal setup side points:591

deleted pawn side points:409

 

Apothecary 2 results:

normal setup side wins: 531

draws:164

deleted pawn side wins:305

normal setup side points:613

deleted pawn side points:387