To clarify, here's what I did and referred to as "forward/backward analysis". I had Cfish-dev search from the root position to some considerable depth. I then stepped forward through Cfish's PV from the root position to the end of the PV. I let Cfish search that terminal position, and it settled on an evaluation of 0.00. I then back-tracked to the root position, allowing Cfish time to find anything better than 0.00 at each step. At the end of this process, Cfish was back at the root position, and a second extended search of that position, with help from the 64 GB hash table, remained at 0.00.Dann Corbit wrote: ↑Tue Aug 06, 2019 6:53 pmwhat you are describing is more like the analysis performed by Aquarium, where lines are explored iteratively.Look wrote: ↑Tue Aug 06, 2019 1:24 pmI am not sure whether I understand your question fully, so you may explain more. But from what I infer, first we have forward analysis until we find an important but neglected move. Then backward analysis and now engine knows about that move to begin with. The cycle goes on till you have a complete image of the initial position.Dann Corbit wrote: ↑Tue Aug 06, 2019 7:56 ambut forward backward analysis assumes the solution. Otherwise, what are you going backwards from?
Perhaps it is my ignorance of the nomenclature, but what I call forwards and backwards analysis involves taking a pre-existing plan/pv/game/ and analysing backwards and forwards over the projected nodes, choosing new nodes when the engine finds improvements,
Ortueta vs Sanz 1933
Moderators: hgm, Rebel, chrisw
-
- Posts: 6442
- Joined: Tue Jan 09, 2007 12:31 am
- Location: PA USA
- Full name: Louis Zulli
Re: Ortueta vs Sanz 1933
-
- Posts: 396
- Joined: Sat May 05, 2012 2:48 pm
- Full name: Oliver Roese
Re: Ortueta vs Sanz 1933
Very good Prof. Zullil, you have described the method exactly.zullil wrote: ↑Tue Aug 06, 2019 7:07 pm ...
To clarify, here's what I did and referred to as "forward/backward analysis". I had Cfish-dev search from the root position to some considerable depth. I then stepped forward through Cfish's PV from the root position to the end of the PV. I let Cfish search that terminal position, and it settled on an evaluation of 0.00. I then back-tracked to the root position, allowing Cfish time to find anything better than 0.00 at each step. At the end of this process, Cfish was back at the root position, and a second extended search of that position, with help from the 64 GB hash table, remained at 0.00.
I have been using it for over 20 years now [sic!] and occasionally tried to discuss it, but typically nobody seemed to care here.
Of course in a computer chess forum, users typically don't enjoy themselves studying chess very much, that could be a problem.
This method happened to work with the earliest versions of Shredder, but works even better with modern versions of stockfish/cfish. Results can be most dramatically in the endgame, if you habe table bases installed.
Of course you can mess it up also: If for example there are two moves which are nearly equal and you explore the first move, fish may stick with it for a long time, even if the second move would have been better in the long run. To overcome this, i may give identically looking moves equal search time.
-
- Posts: 365
- Joined: Thu Jun 05, 2014 2:14 pm
- Location: Iran
- Full name: Mehdi Amini
Re: Ortueta vs Sanz 1933
[...]
Aother point I may mention is that as the engines progress , the quality difference between F/B A and infinite analysis is reduced.
I roughly remember that iterative analysis by Aquarium called IDeA. No it not not same as what was called forward/backward analysis (F/B A) , since F/B A is done in one session, hence zullil's (currently) huge RAM. I remember you could actually save the tree in IDea for future sessions.Dann Corbit wrote: ↑Tue Aug 06, 2019 6:53 pm
what you are describing is more like the analysis performed by Aquarium, where lines are explored iteratively.
Perhaps it is my ignorance of the nomenclature, but what I call forwards and backwards analysis involves taking a pre-existing plan/pv/game/ and analysing backwards and forwards over the projected nodes, choosing new nodes when the engine finds improvements,
Aother point I may mention is that as the engines progress , the quality difference between F/B A and infinite analysis is reduced.
Farewell.