TalkChess.com

Aser Huerga wrote:Any chance you will implement Multi PV in the near future?

I started implementing it last week but realised it required some thinking to do right. So I cancelled. Unfortunately you will have to wait for a possible next version for MultiPV.

Evert wrote:What I find particularly impressive (perhaps you can say something about this) is that the different weights and parameters all look "reasonable" but they don't look like they have been tuned to death. Is this correct?

Hi Evert!

I would find tuning to death very boring. When a feature looks particularly important, we try a couple of alternative weights. But as you know that doesn't take correlation into account. A middle ground.

Xann wrote:It has a brother!

An Explorer Pro bought in a physical shop (an accomplishment these days). I don't like the design, but the screen is marvellous. Any idea why NPS doesn't go up in endgame? I am suspecting a series of random numbers

Collection is going well, two programmers already! But Richard Lang, pretty expensive. So I think I'm going to stop here.

No idea why The Ex Pro dosent get higher NPS

You have a Kittinger and a Morsch

you stlll need a Lang,Schroeder,Spracklen ,Kaplan
and most interestingly a ...

Donninger Regards
Steve

For Fabien:

Very interesting and easy to get dedicated chess computers are

Mephisto III (doing 1-3 NPS) but plays good chess !!!!!!
How the hell did Nitsche and Henne build a commercial chess computer by doing 1-3 NPS ???? in 1983 and even winning a championship title 1984 with the 68000 version of it ?

Mark V (B-Strategy often missing something, from 1981) and the more powerful but still to selective Sphinx 40/50 with Mark V program on 68000 CPU.

several Computers with gyula horvath engine PANDIX:
Krypton Regency/Challenge,
plays very good positional chess.

Xann wrote:Hi all,

I have recently regained interest in computer chess. By a coincidence (or fate?), Joachim Rang and Ryan Benitez are also active once again.

I have been working on a new GPL engine: Senpai.

The highlights compared to Fruit are:
- from-scratch engine (no Fruit code)
- plain bitboards (I will implement magic bitboards later)
- more precise mobility (safety, centre weights)
- evaluation of tactical moves: captures and checks
- late-move pruning (last few plies)
- more aggressive futility pruning
- SMP

The download page: http://www.chessprogramming.net/senpai/
Direct link: http://goo.gl/H0h1Jd
Estimated level: Protector or better (don't ask).

Have fun with Senpai!

Fabien.

Welcome back Fabien. It's nice to see you're doing well. Things just got interesting, I especially love the name although you may need to rename Fruit to "kōhai"!

best,
--tom

looking at the source I see that the search has
const int MAX_PLY = 100;

I think that 100 plies were enough some years ago when the hardware and software were weaker but they are not enough today with good pruning and good extensions.

I believe that with check extensions and singular extensions there may be positions that you need to search more than 100 plies forward in some lines to find a better move.

An interesting test may be to take 100,000 random positions from game and search them for 100,000 nodes with 1 processor to see in how many positions you get a different move when you increase MAX_PLY

You can start with small MAX_PLY and increase MAX_PLY by 1 until you find the minimal value that you see no move change when you increase MAX_PLY

Later you can repeat the process also for 1,000,000 nodes or 10,000,000 nodes and try to get a formula for the maximal ply that you practically need for fixed number of nodes.

After doing it I think that it is better to be careful and have MAX_PLY that is probably enough also for 10^12 nodes so it can be enough also for correspondence games in the near future

You can do the same also for MAX_DEPTH to see if
MAX_DEPTH = 100 is practically enough or you need a bigger number.

Is there any change to see it on github so that we can follow development/submit patches ?

Great to see FL back in action! What ELO strength is Senpai 1.0 starting at? For sure some sort of evaluation tuning would be necessary to give the engine a dramatic boost, but if the internal geography of the engine is good and efficient then the sky is the limit to how high it can go!

j_romang wrote:Is there any change to see it on github so that we can follow development/submit patches ?

That would be the first step. Second step would be to have a public distributed testing framework.

j_romang wrote:Is there any change to see it on github so that we can follow development/submit patches ?

Feel free to send me pull requests:
https://github.com/lucasart/Ninja