TalkChess.com

[Eelco de Groot]

Hello Majd,

To be honest, I am not sure it is either of those at this point It was more like a little fish on the line I suppose. It is mainly intended for long analysis, but the effect is, at least as far as I can see, the opposite of that much more in the first few iterations, and since Stockfish does not produce output anymore in the first second, I'm not quite sure what exactly is happening... To be honest I am mainly just testing at the moment if I can go to depth 28 with it, because this is supposed to be for analysis, but it puts a lot of additional strain on the processor stack so the previous two versions got as deep as 27 but then were never heard of again. There was no actual crash and the taskmanager is still happily going all green on my Q6700 but no more output for 1000 minutes after the 27th iteration were it even managed to do those 27 iterations in just 75 minutes. That 75 minutes was good but I realize those stack overflows will not be good for Uri's nerves, I can just anticipate the disappointment so I had to scale back the extra search at the tips at depth 28 and further, going back to just four plies and I am hoping that this was a typical middlegame position I tested and that my measures can now make the engine go deeper than 28 plies. If I just kept a simple four plies as the extra depth in the last twelve plies of the PV, the stack did seem to hold that but that is not really what I wanted. I want to replace the final position, at the ultimate point of the PV, which in all classical programs is just one position plus quiescence search, and replace that with a search of at minimum four plies as I mentioned but preferably more, I hope for at least ten plies instead of the one eval... But the stack blew up in my face at depth 28... So now I have to go back to that basic four ply search at the tip at that depth. It may not be such a bad thing because the PV is not just the end leaves, the trunk of the tree has be able to carry the crown and keep the tree standing if there is a storm..

Keeping that in mind, to compensate scaling back the PV scout-searches at depth 28 I now try to widen the search again in the trunk of the PV, so the opposite of sharpening the tip. The last build now that does this is at depth 28 and holding, so the game seems on again! Only a few days left to make changes, not that I am entering the tournament but I promised it had to be ready by then, for all the top programs

You asked what "Sharpening of the PV tips" is all about, but the term really is something Mark Uniacke came up with. I don't know if he wrote anything about it, I have not seen it mentioned by Mark or Harvey in the past in the Hiarcs forum for instance? I also have no later versions of Hiarcs than my Hiarcs 12 so I do not know if it is still one of the options you can set today, I think it was optional in Hiarcs 12 but I'm not sure.
I just let my imagination try to come up with something that could do that and it may be theoretically all wrong, what I have, there is no way I can tell without further testing. But I don't think it really hurts either, so as far as I am concerned it can stay in the code.

Thanks for the interest Majd!

Eelco
_________________
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan