From 5 ply to 6....

[fern]

What a big difference an extra ply make! I mean, in this still narrow surface of a game which can be measured by half a dozen ply or less. Perhaps there is not much difference between ply 18 and ply19, but in such a short counting as 5 or 6, it is enormous.
i am talking of old machines. I just had a trip to Buenos Aires and carried with me a hand held unit, Chess Touch de Luxe by Excalibur, with a program buy Ron nelson, one of the father of this industry and in fact lot more famous for that historical reason than for the acumen of his programming.
At first I set the machine to play 40 moves in one hour, my favorite rhythm, in which frame of time the machine examine around 5 ply in middle game. I trounced it easily.
Then I became magnanimous and gave the machine one hour and half for same 40 moves and the contrivance begun to look around 6 moves or more in middle game. And certainly as I kept my carefree approach to the machine, It was me, this time, the trounced side. I could see, after looking the games, that in the same position he played LOT better moves, saw lot more tactical tricks and played a lot more strong game. And just because one extra ply!
I wonder what our programmers here can elaborate about this...


Fern

[JuLieN]

What a big difference an extra ply make! I mean, in this still narrow surface of a game which can be measured by half a dozen ply or less. Perhaps there is not much difference between ply 18 and ply19, but in such a short counting as 5 or 6, it is enormous.
i am talking of old machines. I just had a trip to Buenos Aires and carried with me a hand held unit, Chess Touch de Luxe by Excalibur, with a program buy Ron nelson, one of the father of this industry and in fact lot more famous for that historical reason than for the acumen of his programming.
At first I set the machine to play 40 moves in one hour, my favorite rhythm, in which frame of time the machine examine around 5 ply in middle game. I trounced it easily.
Then I became magnanimous and gave the machine one hour and half for same 40 moves and the contrivance begun to look around 6 moves or more in middle game. And certainly as I kept my carefree approach to the machine, It was me, this time, the trounced side. I could see, after looking the games, that in the same position he played LOT better moves, saw lot more tactical tricks and played a lot more strong game. And just because one extra ply!
I wonder what our programmers here can elaborate about this...


Fern

Going from 18 to 19 plies increases the depth by 5.6%, Fern, whereas going from 5 to 6 plies increases it by 20%, which should then be 3.6 times more noticeable

[fern]

I know that elemental maths, Julien, but surely the thing is somewhat more complicated than that....I suppose the answer is that the nearest the root, the nearest dramatic differences between one move and another. At ply 18 almost everything of the position has been already extracted...Ply 19 probably does not add much...

Fern

[JuLieN]

I know that elemental maths, Julien, but surely the thing is somewhat more complicated than that....I suppose the answer is that the nearest the root, the nearest dramatic differences between one move and another. At ply 18 almost everything of the position has been already extracted...Ply 19 probably does not add much...

Fern

Yes, the more plies a tactic needs, the rarest it statistically becomes. So most of tactics happen in less than 10 plies: above ten plies the engine will rather increase in strategical strength.

Just picture that:

- at 1-ply depth, you can win a chessman your opponent left en prise.
- at 3-plies you can win a chessman with a discovered check or a fork.

So after a while, such tactics don't benefit from deeper searches.

[fern]

Yes, I was thinking along that line. Even so, I have considered the fact that it is always possible that after "n" ply. whatever the number, a dramatic new situation could arise. This is more clear with the so called horizon effect, when few ply are in stake, but it could happens in long lines, as in endings.
Maybe this problen cannot be analysed using simple linear maths or, in common logic, as a regular function going up or down in some accountable way. Some points of sudden irruption of dramatis personae can happen everywhere...

Fern

[JuLieN]

Yes, I was thinking along that line. Even so, I have considered the fact that it is always possible that after "n" ply. whatever the number, a dramatic new situation could arise. This is more clear with the so called horizon effect, when few ply are in stake, but it could happens in long lines, as in endings.
Maybe this problen cannot be analysed using simple linear maths or, in common logic, as a regular function going up or down in some accountable way. Some points of sudden irruption of dramatis personae can happen everywhere...

Fern

Some tactics that could take bigger plies depths would for example involve the suppression of defensive pieces (for example, if the only piece that prevents white from checkmating in h8 with their Rh2 and Rh1 was a Ng6 then playing the Qd3xg6 sacrifice would be part of the tactics. The more defensive pieces you need to capture or divert, the rarest the tactic is.

It would btw be interesting to know what the longest known tactical moves are (excluding checkmate attacks, only accepting gain of material tactics).

So, all this point to one conclusion: increasing the depth of search produces its best tactical results near the root of the search. And as tactic gains are more decisive that slight strategical gains, increasing a shallow search produces more results than increasing a yet deep search.

[fern]

There is another and very known situation that happened a lot to older engines in endings, that is, situations that are decisive without captures, as when you can get a queen with a pawn after a race for it and some perhaps intermediate moves. There you have lines beyond 10 ply and the end of them is decisive without shedding of blood.
Maybe in the really high level of chess, that played between SGM, is like that: after extremely long lines, full of complications, both players understand that something dramatic has been reached and one of them surrender and we, spectators, look in amazement all that because we do not see that very far conclusion.
No fork, not piece taken, but a decisive output all the same..

Fern

[JuLieN]

There is another and very known situation that happened a lot to older engines in endings, that is, situations that are decisive without captures, as when you can get a queen with a pawn after a race for it and some perhaps intermediate moves. There you have lines beyond 10 ply and the end of them is decisive without shedding of blood.
Maybe in the really high level of chess, that played between SGM, is like that: after extremely long lines, full of complications, both players understand that something dramatic has been reached and one of them surrender and we, spectators, look in amazement all that because we do not see that very far conclusion.
No fork, not piece taken, but a decisive output all the same..

Fern

Yes, but understand me well, Fernando. Those very deep lines are rare. You get like 99% chances to lose a game against a computer because it grabbed a pawn during a 3-plies search, and then a knight in 3 plies too, and so on, than losing a deep strategical game against it (except if you are a GM). So the more of those 99% tactics the computer can get, the stronger it will become against a human. I bet your pocket game jumped from, say, 60% of getting those tactics to maybe 80% just by becoming able to search one ply deeper, hence the big increase in strength you noticed.

[JuLieN]

I'll rephrase it. If 99% of tactics can happen during, say, a 10-plies search, then adding 3 extra-plies won't increase its tactical force because 10 plies are enough to find 99% of tactics. So the 3 extra plies will rather increase its strategical strength, which benefits less from a new added ply than tactics near the root.

[fern]

I understand you well, Julien, I feel all that. No mystery. Normal games are won or lost in that way at our level of players not better than, say,2200 or so.

In fact I have few doubts about this, my goal was more to express the feeling of playing with the same machine at an slightly longer rhythm and the effect of it that to begin a scholastic thread about search trees and all that. .

It is a sobering experience because no matter how much better you are, theoretically, compared to that hand held unit or any other machine or engine, it still can beat you in that tactical level if you lose a little bit your alertness.

a lesson for us all: always the turtle can beat the rabbit if this last does not push his strength to the full.

rabbit fern