Only alphabeta engine

[OttoLau]

Has anyone ever tried the limits of how strong of an engine you can get only using safe pruning methods and their enhancements?
Talking about:
Alphabeta
Mate distance pruning
PVS
TT
Aspiration windows
Good move ordering
And not using any unsafe pruning:
Razoring
NMP
LMR
Delta inside quiescence
etc.
Would be intresting to know the effect of full knowledge search versus the deeper but less accurate pruned search

[abulmo2]

Has anyone ever tried the limits of how strong of an engine you can get only using safe pruning methods and their enhancements?
Talking about:
Alphabeta
Mate distance pruning
PVS
TT
Aspiration windows
Good move ordering
And not using any unsafe pruning:
Razoring
NMP
LMR
Delta inside quiescence
etc.
Would be intresting to know the effect of full knowledge search versus the deeper but less accurate pruned search

In my simple engine Dumb, unsafe pruning brings about 450 Elo. Most of the pruned lines are actually bad, so it is worth discarding them. More precisely I got:
* Razoring: 50 Elo
* Futility pruning: 18 Elo
* Null move pruning: 92 Elo
* probcut : 8 Elo
* Late move reduction : 259 Elo
* Late move pruning : 27 Elo
Of course, move ordering and accuracy of the evaluation function also play a part to this Elo improvement.

[syzygy]

Has anyone ever tried the limits of how strong of an engine you can get only using safe pruning methods and their enhancements?
Talking about:
Alphabeta
Mate distance pruning
PVS
TT
Aspiration windows
Good move ordering
And not using any unsafe pruning:
Razoring
NMP
LMR
Delta inside quiescence
etc.
Would be intresting to know the effect of full knowledge search versus the deeper but less accurate pruned search

See Deep Blue. You need 200Mnps to achieve around 2700 Elo.

[Uri Blass]

Has anyone ever tried the limits of how strong of an engine you can get only using safe pruning methods and their enhancements?
Talking about:
Alphabeta
Mate distance pruning
PVS
TT
Aspiration windows
Good move ordering
And not using any unsafe pruning:
Razoring
NMP
LMR
Delta inside quiescence
etc.
Would be intresting to know the effect of full knowledge search versus the deeper but less accurate pruned search

See Deep Blue. You need 200Mnps to achieve around 2700 Elo.

Deep blue does not use NNUE evaluation. I guess that with NNUE evaluation that does not contradict only using alpha beta in search you can get a lot more than Deep Blue.

[syzygy]

Has anyone ever tried the limits of how strong of an engine you can get only using safe pruning methods and their enhancements?
Talking about:
Alphabeta
Mate distance pruning
PVS
TT
Aspiration windows
Good move ordering
And not using any unsafe pruning:
Razoring
NMP
LMR
Delta inside quiescence
etc.
Would be intresting to know the effect of full knowledge search versus the deeper but less accurate pruned search

See Deep Blue. You need 200Mnps to achieve around 2700 Elo.

Deep blue does not use NNUE evaluation. I guess that with NNUE evaluation that does not contradict only using alpha beta in search you can get a lot more than Deep Blue.

True, but we know that pre-NNUE versions of SF were already much stronger than Deep Blue (at much lower nps). But I will admit that it is not really fair to point to a nearly 30-year old "engine".

The Deep Blue creators decided against non-safe pruning methods (I think because its predecessor Deep Thought once missed a tactic). They will not have realized how much they crippled Deep Blue with that decision.

I once removed what I think are exactly the non-safe pruning methods from Stockfish(/Cfish) to create Cripplefish:
https://github.com/syzygy1/Cfish/commit ... 254428c6b0
Cripplefish tested 800 Elo weaker.

[OttoLau]

Has anyone ever tried the limits of how strong of an engine you can get only using safe pruning methods and their enhancements?
Talking about:
Alphabeta
Mate distance pruning
PVS
TT
Aspiration windows
Good move ordering
And not using any unsafe pruning:
Razoring
NMP
LMR
Delta inside quiescence
etc.
Would be intresting to know the effect of full knowledge search versus the deeper but less accurate pruned search

In my simple engine Dumb, unsafe pruning brings about 450 Elo. Most of the pruned lines are actually bad, so it is worth discarding them. More precisely I got:
* Razoring: 50 Elo
* Futility pruning: 18 Elo
* Null move pruning: 92 Elo
* probcut : 8 Elo
* Late move reduction : 259 Elo
* Late move pruning : 27 Elo
Of course, move ordering and accuracy of the evaluation function also play a part to this Elo improvement.

Hello Richard
Good info, kind of weird that probcut is only 8 elo, futility only 18? Then LMR is massive 259.
Perhaps only using safe prunings would be better for an engine that is specialized in solving puzzles?

[OttoLau]

Has anyone ever tried the limits of how strong of an engine you can get only using safe pruning methods and their enhancements?
Talking about:
Alphabeta
Mate distance pruning
PVS
TT
Aspiration windows
Good move ordering
And not using any unsafe pruning:
Razoring
NMP
LMR
Delta inside quiescence
etc.
Would be intresting to know the effect of full knowledge search versus the deeper but less accurate pruned search

See Deep Blue. You need 200Mnps to achieve around 2700 Elo.

Deep blue really didnt use any unsafe ones?
Is that why it didnt reach that high depths despite so much nps?

[Bo Persson]

The Deep Blue creators decided against non-safe pruning methods (I think because its predecessor Deep Thought once missed a tactic). They will not have realized how much they crippled Deep Blue with that decision.

They did reach the project's goal of beating the world champion. Mission accomplished!

The project was under time pressure by the custom chips arriving late, so had limited time for the final tests. At that point, going for Good Enough seems like a reasonable idea.

(Had this not been a marketing stunt by IBM, they could have continued the development after the match. However, IBM was satisifed with being #1, and retired the hardware).

[syzygy]

Has anyone ever tried the limits of how strong of an engine you can get only using safe pruning methods and their enhancements?
Talking about:
Alphabeta
Mate distance pruning
PVS
TT
Aspiration windows
Good move ordering
And not using any unsafe pruning:
Razoring
NMP
LMR
Delta inside quiescence
etc.
Would be intresting to know the effect of full knowledge search versus the deeper but less accurate pruned search

See Deep Blue. You need 200Mnps to achieve around 2700 Elo.

Deep blue really didnt use any unsafe ones?
Is that why it didnt reach that high depths despite so much nps?

Yes and yes. Not even nullmove. Just full width alpha-beta plus extensions.

https://sjeng.org/ftp/deepblue.pdf

The search should provide "insurance" against simple errors. We wanted to be sure that all move sequences were explored to some reasonable minimum depth. Early research into pruning algorithms (e.g., null move pruning [3, 9]) did not provide us enough evidence to warrant implementation in the hardware search of Deep Thought 2 or Deep Blue. Even without pruning, and using highly selective search, we felt that Deep Blue had suffcient searching power to satisfy our insurance needs. A three minute search on Deep Blue would reach a full-width depth of 12.2 on average.

So by "highly selective" they mean extensions.
The hardware apparently did support nullmove search, but they never used it (see section 3.4).

Full-width depth of 12.2 (plus extensions) is of course stronger than a 12-ply selective search, but apparently 2-3 ply extra already compensates for that:
viewtopic.php?p=747232#p747232

For the curious, if you play Cripplefish with Cfish at fixed depth, Cripplefish will be much stronger. I tried some time ago with Andscacs and if I remember well at depth 10 was maybe 200 elo stronger. But only that you play Cripplefish at depth 10 and Cfish at depth 12-13, Cfish will win.
I tried also at higher depths, and the difference of strength at the same depth diminishes.

Of course if Deep Blue did many more extensions than modern engines, this could somewhat compensate for the lack of reductions and pruning. Still, it is remarkable that Deep Blue was strong enough to give Kasparov a difficult time.

[syzygy]

The Deep Blue creators decided against non-safe pruning methods (I think because its predecessor Deep Thought once missed a tactic). They will not have realized how much they crippled Deep Blue with that decision.

They did reach the project's goal of beating the world champion. Mission accomplished!

The project was under time pressure by the custom chips arriving late, so had limited time for the final tests. At that point, going for Good Enough seems like a reasonable idea.

(Had this not been a marketing stunt by IBM, they could have continued the development after the match. However, IBM was satisifed with being #1, and retired the hardware).

There was one year between 1996 and 1997. But they "concluded that the searching ability of Deep Blue was acceptable, and [they] spent the vast majority of our time between the two matches designing, testing, and tuning the new evaluation function".

I'm not saying they made poor decisions with the knowledge they had. But with the knowledge we have now, they could probably have crushed Kasparov Lee Sedol-style if they had implemented a more "modern" search. (But there was no Stockfish or Fruit/Ryba/Ippolit to take ideas from, yet. So it would probably have been just nullmove.)