Moderator: Ras
Le Damier de l'Opéra? Yes they're still here.fern wrote:Colossus by Bryant was strong. I had Colossus 4 running on Atari, loaded from a tape....It usually bested my beloved Champion Fidelity, but not always. Once, in Paris, I went from one shop to another to get Colossus X....Those were the days.
I hope the old place where I got every thing chessy in Paris is still there...near l'Opera....
Fern
Hmm... but if you look at the game, you'll see that Sargon won for obvious positional reasons. That's why it displayed a >1 score loooong before any material advantage was computablefern wrote:There is more than evaluation in argon II; it search full width....
Maybe you should the same for a time. The, slowly, you prune here and there with some evaluations tricks. Full width avoid lots of problems and with fast processor, yous till can get perhaps 7 ply or more.
Minimal pruning regards
Fern.
My current beta of Prédateur is very conservative, regarding pruning. it prunes a lot less than the previous versions (and plays stronger). Pruning, when well done (for instance with Null Move), allows the engine to catch deeper tactical menaces. So, nowadays, such a "full - width" engine would just be crushed tactically, actually.fern wrote:One idea I have had since long time ago is this: a processor divided in two modules. The first do full width search to see obvious tactical tricks; if nothing important is got here, second module go ahead with pruning and the rest to see for positional considerations.
What do you think?
F
Something similar to this was tried in the past. In the late '70s, the old MacHack program was coupled with Cheops, a brute force tactical searcher that would override the moves of the slower, more positional MacHack if it found that the move MacHack chose had a tactical hole in it. Nowadays, people use multiple processors merely to speed things up. I wonder if this idea would still be viable today; it might be interesting to revisit.JuLieN wrote:
Pruning, when well done (for instance with Null Move), allows the engine to catch deeper tactical menaces. So, nowadays, such a "full - width" engine would just be crushed tactically, actually.
Now, another idea that we all more or less thought about one day, is having a tactical module and a positional module, that would go that way:
1- the tactical module makes a search and look if there is a clear winning move (mate sequence or material winning). If yes, play this move.
2- if not, let the positional module propose a move.
3- test the proposed move using the tactical module: if it leads to a material disaster, go back to 2; waiting for another proposal from the positional module. If not, play this move.
But all in all, having a good evaluation function assessing the position on the leaves of the tree actually does that already.
That made sense with MacHack (MH), because MH used a plausible move generator: it only generated a handful of moves it thought were promising (the closer to the leaves, the less moves), and hence was subject to tactical mistakes. I don't think using plausible moves generators would be very popular nowadaystmokonen wrote:Something similar to this was tried in the past. In the late '70s, the old MacHack program was coupled with Cheops, a brute force tactical searcher that would override the moves of the slower, more positional MacHack if it found that the move MacHack chose had a tactical hole in it. Nowadays, people use multiple processors merely to speed things up. I wonder if this idea would still be viable today; it might be interesting to revisit.JuLieN wrote:
Pruning, when well done (for instance with Null Move), allows the engine to catch deeper tactical menaces. So, nowadays, such a "full - width" engine would just be crushed tactically, actually.
Now, another idea that we all more or less thought about one day, is having a tactical module and a positional module, that would go that way:
1- the tactical module makes a search and look if there is a clear winning move (mate sequence or material winning). If yes, play this move.
2- if not, let the positional module propose a move.
3- test the proposed move using the tactical module: if it leads to a material disaster, go back to 2; waiting for another proposal from the positional module. If not, play this move.
But all in all, having a good evaluation function assessing the position on the leaves of the tree actually does that already.