"Full-width means never having to say you're sorry"

[fern]

Maybe we are now in an epoch of hardware development when compromises are not anymore needed. In theory, to look at every possible move is the best approach, as much by definition you cannot miss something, but of course was not that good and feasible in ages when full width search made possible only to go 4-5 ply deep in a decent interval of time. So prunning was the answer because missing something was less harmful that not looking at something at all, due to ply limitation. Since that principle existed, many innovations followed: the one by land, null move, etc. Techniques to make more effective the materialization of the in principle faulty idea of just discard some moves on presumption they were not useful.
but now? The processing power exist more and more to make possible deep width search. I wonder who will be the guy that this time will do as Lang do, but now with the opposite principle....

My best
fernando

[Kanizsa]

similar scenes appeared in the software times. i remember endless debates with ossi weiner about the accurate genius method of computing.
in relation to the "inaccurate" and speculative method of Marty Hirsch's Mchess.

At that time (1993-1995) I think that Lang's programs were better, but Marty take the mission to defeat Genius, at the cost to unbalance the style.

It was evident to all that M Chess was weaker in rapid times or equal hardware, but Marty was confident that introducing more knowledge during many and many updates the next update should be slow than the previous one, but could take benefit from the rising hardware.
Marty was very confident in the law of Moore. In those years the new AMD K5 and K6 appeared, very better than Intel CPU in crunching chess numbers, and so Marty was able to overcome Genius 5 (running on Intel) definitevely, at Paderborn.

The effort to overcome Genius was though for Marty. For instance, the unrealistic evaluation to give dinamic contempt agains an ultradefender like Genius. In last versions, MChess was tipically showing to +1,00-1,50 in equal positions.

The program becomed too tailored vs. Genius and so inadeguate to contrast the newcomers, in primis Fritz and Shredder, without radical rifondation and rewriting.

So ended the beautiful story of Marty Hirsch and MChess....

[mclane]

i see it a little different.
there is no progress in genius 2-3-4-5-6.
in the same way there is no progress in shredder 7-8-9-10...
marty gave up because he was betrayed by the dealers.
he had to make a living from programming other stuff.
i am sure he would / could come back and show us that he can compete with anybody. he is a very clever guy.
i really like him and his program.
we all have great respect for marty hirschs work.

[Kanizsa]

i really like him and his program.
we all have great respect for marty hirschs work.

I also too, but the unrealistic positive evaluations of M Chess was the price to pay to overcome some passive programs, like Genius was.
In last versions, MChess tipically showed up to +1,00-1,50 in equal positions. This positive contempt became inadeguate to face the newcomers, very skilled in tactics due to the null move.

[PK]

It's funny that recently I am also trying to unbalance my program in order to get better results (and the playing style more pleasant to look at) - the only difference being that it is done at the level of about 2000-2100 Elo, and not at the top. Thank You for pointing a historical example that it actually might work to some extent.

General direction is to emphasize king attack for both sides and mobility for the opponent (so that program attacks and simultaneously avoids positions where opponent has attack potential)

My problem with this approach is that the reduction techniques which were sort of OK for a more balanced program don't seem to work any more. Is it a common experience?

regards from an avid reader of Thorsten's posts

[tiger]

Maybe we are now in an epoch of hardware development when compromises are not anymore needed. In theory, to look at every possible move is the best approach, as much by definition you cannot miss something, but of course was not that good and feasible in ages when full width search made possible only to go 4-5 ply deep in a decent interval of time. So prunning was the answer because missing something was less harmful that not looking at something at all, due to ply limitation. Since that principle existed, many innovations followed: the one by land, null move, etc. Techniques to make more effective the materialization of the in principle faulty idea of just discard some moves on presumption they were not useful.
but now? The processing power exist more and more to make possible deep width search. I wonder who will be the guy that this time will do as Lang do, but now with the opposite principle....

My best
fernando

It will not happen. Successful selective techniques scale very well with more processor power, so brute force is definitely dead.

In other words, if your selective search works better than brute force at ply depth 3 or 4, it will also work better at ply depth 10, 15, 20 and 30, and generally it will work exponentially better at those higher depths.

You can forget about brute force.


// Christophe

[fern]

Ok, so it be, you are the expert
How is going on ...
ton famille?
Et certaimement, quand es que nous verons un nouveau Tiger?

santé
Ferdinan

[tiger]

Ok, so it be, you are the expert
How is going on ...
ton famille?
Et certaimement, quand es que nous verons un nouveau Tiger?

santé
Ferdinan

Playing with my 3 years old daughter at this time. Well... I cannot sit in front of my computer. Now she's calling me... again.


// Christophe

[Dann Corbit]

The main gist of that article (as I recall) was that the former attempts at making the best move rather than full width search did not fare as well as full width searching.

Some of the earliest chess programs would analyze the spots off of a move and then make that one without searching all of the alternatives. It was a pretty good strategy when you could only search a few hundred nodes per second. When CPUs got faster, full width searching became practical.

Of course, the value of null move and history reductions is clear now (which is not an elimination of full width search, but a delaying of it in a sense, because if you search deep enough you will eventually search everything anyway).

So we can change the quote to:
"Full-width means never having to say you're sorry, but you don't get to say anything until a week from next Thursday. Better do some pruning, or that shrub will turn out to be 'General Sherman'."

[mongrel]

(Christophe Theron, you have a PM about a book, Sargon I, by the Spracklens.)