Computer Chess Club Archives


Search

Terms

Messages

Subject: pmess at cct

Author: Mridul Muralidharan

Date: 07:39:25 02/27/06


Hi,

  Here is a short writeup of the games I observed ....

I seem to have unfortunately lost the logs from the first day ... so 'game
comments' from memory.
The tourny started in the stupidist way possible - I overslept and missed the
first round !

R1 : Pmess vs Hossa
Forfiet. Sorry for not being able to make it Steffen - alarm clock went bonkers
:-(

R2 : Tornado vs Pmess
Pmess playing without a book - and always falls for the stupid opening line
(imho)

1. d4 d5 2. c4 e6 3. Nc3 Bb4 4. Nf3 dxc4 5. e4 (or e3 for that matter) b5?

It thinks it can retain the pawn (with slim margin of +0.1 or so) , keeps
playing - and keeps messing up its position !
Luckily , tornado did not convert this advantage into a win and (as you will see
later) inspite of some bugs , pmess managed to win this one.

R3 : Arasan vs Pmess

Arasan has traditionally been a problem engine for most of my engines.
I had not tested the last couple of them against it , so had no idea what to
expect.
It was a similar opening as r2 , and you will see the similar b5 move ... though
I think it was not as stupid in the game as earlier on - comes sufficiently
later to be not losing too much tempi (but it is still quite bad).
By middle game , arasan 'owned' the center and I was expecting a crushing
defeat.
My suspicion about why arasan did not win this match was that it was a bit too
protective of its king side pawns and secondly (maybe more importantly) 'cos it
created holes in the center by pawn pushes - allowing the bishop in a6/b7 to get
back into the game.
According to pmess , around move 28 or so the game started slipping away from
arasan (score became +ve for itself at around that time).
[D]rr6/1bp3qk/1n2pn1p/5p2/2pP1P2/B1P2N2/4B1PP/1RQ2RK1 w - - 0 29
As you can see here , pmess has a lousy pawn structure - a definition of why you
need good pawn structure eval code ! (I added it to my todo list to revamp that
code :-) ).
A really tense game - once again highlighting the 'need' for a opening book.

Another interesting position is this one -
[D]rr6/2p4k/4p2p/3n1p2/2pPbPq1/B1P5/1R4PP/2Q1NRK1 b - - 0 33
pmess failed high on Rxa3 (it was thinking of Rg8 until then).

All in all , a tough fight against a tough opponent.

R4 : Pmess vs Chepla

I have been playing some sparring matches against chepla before cct and result
has been roughly equal (with chepla with its monster nps having a lead :-) ).
Chepla gets insane depths and nps ... though Riverman slowed it down with added
mobility seeing the way it was playing pre-cct (damn me for pointing this to him
:-) ).
AND , this was the first game where a severe bug finally decided to rear its
head.
The opening was , surprisingly , not too bad for pmess imo (and I am a patzer ,
so dont take my word for this) ... and into the middle game , like Riverman put
it , "pmess owned the center , the king side and the queen side" :-P ofcourse he
was jesting :)
Which is when pmess decided to trade rook for a bishop ... I did fix a bug there
afterwards , but that was not severe enough a problem I guess.
But yes , this was the begining of the downfall for pmess ... from then on , the
score never really went up.
It was a tough seesaw since ... with chepla and pmess alternating in some
maneuvering for control. The e file was a monster for chepla and eventually won
the game for it.
[D]5rk1/1p2q2p/2ppr3/p4pp1/2PQ4/P1N3P1/1P5P/3R1B1K w - - 0 28
According to chepla , this was the begining of the end for pmess ... I did find
a pretty severe bug in my queen eval - and the result was Qa7 here ... but I am
not fully convinced that this was a fatal mistake.
But the resulting moves were definitely impacted by the bug ... not sure which
move had a spike in score fall , but it was a steady decline.
A well deserved win for chepla and spanks for pmess :-)

r5: Postmodernist vs Pmess
This game , by far , had the most severe bug in the whole tourny for pmess.
A crushing kingsafety eval bug which increased the eval by more than a pawn
(more in some others) !
As can be expected , pmess lost !
The relevent position where this is most clearly reflected in the static eval
would be this position (from game) :
[D]5rk1/pp1b1ppp/8/q2n2B1/2Q2p2/b1PB4/P1PK2PP/1R2R3 b - - 0 17
pmess evaluated this as 1.5 in blacks favour ! (after fixing , about 0.4 - 0.5
in blacks favour).
This was clearly reflected in the score's of postmoderninst and pmess in the
game - until pmess began to 'lose' it , the difference was as high as 1 pawn !
As Andrew put it , "Postmodernist is getting bored of pmess's attack" :-)

After an exhausting first day of cct (ya , I stayed up the whole 5 rounds, and I
am in India !) , no sleep for the 'bugged' , I was sitting and fixing my bugs
for the next 3 hours.
I ifdef'ed all my changes (#ifdef CCT_DAY1_CHANGES) so that in case I break
something , I can always recompile with this flag removed :-)
Luckily , it seems that I can code for 36 hours straight ... so nothing broke
:-D
Revamped kingsafety totally , changed some passer eval code (I suspect this was
not that well tested :-( ... see the chessthinkerx game for more comments) and
the center control code - that arasan game was scary !

Considering that I had work deliverables also to complete , cct was really
stretching my time, but made it to day 2 without too much problems ... and no r6
forfiet from me :-)

R6 Tinker vs Pmess
I cobbled up a book from some pgn's lying around to not lose too much time in
the opening ... and what a blunder that was :-(
Came off almost a pawn down from the opening itself ... rest of the match was
one looong struggle to overcome this severe handicap.
I think tinker might have some bugs with draw detection ... since quite a few
times in the game , it was allowing repetition for pmess inspite of showing +ve
eval for itself.
This match ended in a draw.

R7 Pmess vs Thinker
I promptly kicked out the book before this match :-)
I prefer loss due to my messup's than some random pgn book move issues !
This was a surprising game in many ways since I expected to be crushed.
As soon as I saw the pairings , I remember asking Peter whether it was for real
- chessthinker was way too strong to face pmess after it has lost like 3 matches
!
I am still analyzing this game , but my impression while watching this game was
that thinker won due to superior endgame knowledge (this is pathetically low in
pmess) ... pmess was a pawn up till the late-middlegame before I left for
dinner. Came back to find a loss - ah well , better a 'fixable' endgame loss
than a more tougher middle game crush :-)

R8 Pmess vs K9
I really cant do justice to this game - I slept off after starting the game.
Woke up around 1.5 hours later to find a win ...
When going through the game later , I found a very strong king side attack by
pmess that lead to a won rooks vs queen endgame.

R9 Prophet vs Pmess

Same as R8 for this game too ... 6 hours of sleep in 3 days was begining to show
and I slept off after starting this game too (more on that below).
Pmess seemed to have let its queen get 'trapped' in the wrong side of the board
for quite sometime until this position
[D]5r2/p2b2pk/1pn1p2p/3pP1p1/P1pP4/2P1NrP1/1qPBQP1P/R4RK1 b - - 0 33
where it play Nxd4
It was not a staight win after this move though.
Pmess tried to force a draw by 3rep at around move 50 or so, but prophet will
not play ball !
[D]8/R1B3pk/1p4bp/3pp1p1/8/2P1r1Pq/2QNrP1P/5RK1 w - - 0 53
It played Qc1 and pmess eval kept rising since them.
I think prophet might have played some bad moves since that (time pressure ?) ,
the game was over in about 3 moves hence (pmess announced mate score).


I continued to sleep , unaware of what happened in R9 ... and since that was the
last game of the tourny , I did not keep an alarm - but an IM from a friend woke
me up later, and I find that there is a new tourny about to begin : I dont think
this 'cct blitz' was announced earlier on.
Note - it was about to begin : lot of games were not yet started.
Asked peter if I could join and he flat out refused saying it was 'too late'.
Oh well , pmess usually sucks at blitz and might have got crushed anyway :-)

Said goodbye and dragged myself to the bed to get some much needed sleep !
Thanks to all at CCT , was great fun - as always :-) The discussions on 64 were
enlightening , and I think we taught erdo how to use the cli quite well ;-)

What I learned from it ?
* Keep alarms properly , and use two alarm clocks for reduncency !
* Parallel search worked beautifully , my tests on the pgn later on showed a
consistent speedup of >= 1.87 (worst was 1.87)
* My eval approach is not too bad , just buggy like hell :-)
* Move ordering stats were pretty good.
* Sleep better the day before the tourny - similar to how you prepare for an
exam I guess.


Thanks,
Mridul

PS : A bit too long post , should have posted in some site I guess ....



This page took 0.06 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.