Monday, February 14, 2005

CCT7 Day 2

Atlast it is over ... and I cant take anymore of it :)
Now I know what the expression "dead tired" means ... note , I just woke up from a 10 hour sleep (4:30 now ;) ) and still I am reeling !

Ok , back to details on the tourny !
After the all nighter for day 1 , I did not sleep ... except for small fits of drowziness in front of the keyboard.
All the "errors" I saw in the games , I was attempting to fix ... or atleast mitigate : like not developing minors sufficiently enough , small pawn structure issues , AND endgame :)
I managed to massively revamp endgame evaluation .. and it plays quiet well in most endgames now - long way still left to go , but this was quiet an improvement from day1 - like the draw against Ikarus.
Rest of the 3 hours before tourny was in testing - this was when I had food , etc.

First game was Chompster - CEng.
This program is written in java - no kidding ! , and was running on (athlon 2500xp @3200, 1gb RAM). Ofcourse , you use -server option, get everything JIT compiled , use almost no garbase collection (no dynamic object creation , etc) and it would be quiet close to native programs :)
This game began disasterously - opening is a very tricky business : a mistake in opening will come back to bite you 25 moves down the game : and no program can see more than 8 - 9 moves (which is 16 - 18 plies or half moves).
The mistakes that CEng made in opening were pretty massive.
I focussed more on endgame tests and did not focus on opening testing much :(
It quickly lost a pawn , and soon the position - the game was heading for a quick loss.
And then the search and eval of CEng kicked in ... and it battled on : all the while keeping me on the edge of the seat - every eval difference , I wil be cheering or cursing.
The battle went on , and the crux was a knight which was never developed - which almost cost me the game !
And then finally - breakthrough !
Better extensions , better branching factor and most importantly singular extensions saved my ass :)
It saw the draw and went for it , while chompster was still thinking it was winning quiet well : the game ended in a forced draw :)
Later on , Scott Farrell told me that lot of programs were still evaluating that position as win for chompster - 3 moves after CEng said it was a draw : singular extensions rock ! :)

After that killer of a game , I got some 5 mins to pump up some params and tune down some others - in the hope that it will be good : ZERO testing time , only intution based changes ;)

Next match : CEng vs Altamax.
Altamax was running on a Pentium 4 - 2.4 Ghz.
The first move gave me a scare - CEng started with Nf3.
Not a bad move , but CEng usually opens only with e4 or d4 ... so was some tuning not correct ? Am I going to lose ?
I knew nothing about this opponent - most others I have heard of / played against / have vague idea of their strenght.
But luckily , the opening progressed without much hiccups ... I was noting down lines and possible score correction on a book like mad anyway :) - for furthur tuning ;)
This was the first game where CEng was totally in command.
It gained a massive space advantage - hemming all of Altamax's pieces in the 7th and 8th ranks , and then closed in on the kill by busting open its kingside with a massive attack.
I was quiet happy with this game , though I identified some scope for improvement .. and made those changes.

Next was AvernoX vs CEng.
AvernoX was running on AMD Athlon 1.3 Ghz - so slower than mine.
The opening was a Petrov classical attack , and in the 5th move itself CEng bungled playing Qe7 instead of d5.
Vincent told me later that this is a common computer program mistake - one more reason why you _need_ an opening book : the theory is way too vast to programmatically evaluate or find by search depth !
The game started going downhill from here ...
I was thinking dark throughts of what I could have done , changed , etc like
"I should add actual mobility evaluation not the scaled down POS version I have now" , etc , etc :D
It was a tough match - the toughest I ever had in the tourny , and on top of it , my net connection was breaking ... crazy combination and I was completely in pieces.
Half way through , I just gave up and went to watch some tv - could not take the misery anymore :) - trust me , the position was an absolute loss - the eval went as high as 3 pawns in favor of AvernoX I think !
But unlike its stupid creator , CEng merrily went on playing , consolidating , exchanging , improving and in the end - managed to break even and started improving !
I was shocked when I saw this ! It actually went from a 3 pawn deficiet to a full pawn advantage !!
But that was just lack of EGTB speaking and the position soon ended in a drawn endgame - two queens only. No EGTB , so CEng saw no draw.
IM Jon Schroer joked saying that CEng's pieces were better placed :P when he saw that the programs were not calling it a draw !
n the end , the TD ame and arbitrated the game as draw.
Thats it , from now on I will never faith in my program :)
More Calvin and chocolate to you CEng :D

I tried tuning some more params before this game - especially centralisation code to avoid mistakes like in last game , and Nf3 in the game before , and some other params.

The last game (phew !) was CEng - Arasan (was running on Pentium IV 3.06Ghz).
I was dog tired by now - almost 40 hours since I slept - not dozing off 'cos of excessive tiredness in front of the comp , but proper sleep.
That too in a very intense atmosphere , and when the game is not going on , I am coding , checking , bugfixing and testng like a nut !
And so begane the last game - and started with giving me a heart attack.
The variations that CEng was thinking of was all crazy and haywire !
Some param was conflictin with another ... I was (yet again ? :) ) cursing myself.
The worst thing that can happen is to know that a last minute change screwed up a potential win :(
But luckily , the score's ended up stabilising on positions/moves that were actually good ... but it was all touch and go.
Move 10. Ne5 was commented upon as a "human" move :)
And then CEng played a novelty (21. c4! ) and actually saw the result of the move.
It was an absolute beauty of a move and Arasan misjudged it and its eval rose thinking it was gonna gain ... but realised its mistake 2 - 4 moves later , while CEng became happier and happier.
Soon it took a very commanding position and its eval went up as high as 6 pawns in just under 30 moves ... very much in middlegame.
Quiet soon , Arasan threw the towel and resigned when its score was consistently at 6 pawns in CEng's favour for 3 moves or so.

And thus ended CCT7 and CEng : 4 wins , 3 draw , 2 loss.
It stood a modest 13th out of the 44 participants based on Buchholz and 5th based on points.
45 hours without sleep , no book , no EGTB , 2 weekends for the engine (3 if you count this one too) - damn good I say :)

And 10 hours later , I am still tired and sleepy ... anyone coming for breakfast/lunch/snacks ? :)

I will put up the games (and maybe it will contain the comments/eval's also) later on somewhere and try to post a link here ... IF I DONT FORGET :)


Post a Comment

Links to this post:

Create a Link

<< Home