![[de]](/mwf/v12/flags/de.png "Germany")
> I am sure we all waiting for the new Rykba engine to make its comeback real soon. Houdini, be afraid, be very afraid...
I'm afraid!
I'm beginning to be more and more afraid that we are riding a dead horse since we began to speculate about R5 and the further Rybka-development at all.
Quap
> we are riding a dead horse
That is a non sequitur! If a horse is dead- one cannot presume to be riding it.
![[at]](/mwf/v12/flags/at.png "Austria")
He means we are flogging or beating a dead fish.
Rybka 4.1 is very much in play regardless of its strength. It still has viability as an analysis tool. To put it in material terms, it is very much alive. Rybka 5 is yet to be. Rather presumptuous to assume something is dead before it comes into play.
Vas is working on chess engines. Something will pop out sooner or later. I wouldn't bet against him...
![[us]](/mwf/v12/flags/us.png "United States")
Well Said
Thumps Up
Thumps Up
![[ca]](/mwf/v12/flags/ca.png "Canada")
I've been competing with R3 & R4 for a few years for online correspondence and ended up being in the top 5 before retiring (got bored)...I've recently tried Houdini and in my opinion R4.1 is stronger...at least at very long analysis (I then used inifinite and let it run for a few hrs between each move). Houdini doesn't see what R4.1 does as fast and if I were to feed a Houdini move into the same position and let R4.1 analyze then R4.1 would often see it with a significantly lower score. Both engines would reach the same depth for this test.
I also don't like the fact that Houdini needs to be stopped to save the hash. I like to be able to save my hash periodically during the analysis (in case my system crashes or power goes out, etc.). Stopping and restarting Houdini and analyzing from the same position (even without reloading the engine) forces it to analyze from scratch again for some reason (with or without using Preserve Analysis.
I also don't like the fact that Houdini needs to be stopped to save the hash. I like to be able to save my hash periodically during the analysis (in case my system crashes or power goes out, etc.). Stopping and restarting Houdini and analyzing from the same position (even without reloading the engine) forces it to analyze from scratch again for some reason (with or without using Preserve Analysis.
![[mx]](/mwf/v12/flags/mx.png "Mexico")
> I also don't like the fact that Houdini needs to be stopped to save the hash.
Use the other parameter that does that, "Learning" I think it's called, should save automatically.
What exactly does it save automatically? The hash file for that game (the one where I type in the filename) or just the engine hash? I use Aquarium. I still don't like interrupting the engine by stopping it just to save the game's hash (even if it were to start from the same depth)...sometimes the engine is in deep and stopping/starting it again (even if from the same depth) would force it to reanalyze the process it was in which it could've been analyzing for hours already.
> What exactly does it save automatically?
The evaluations of the best moves, so that they don't need to be researched.
> I still don't like interrupting the engine by stopping it just to save the game's hash
You don't need to do that with the Learning parameter.
> sometimes the engine is in deep and stopping/starting it again (even if from the same depth) would force it to reanalyze the process it was in which it could've been analyzing for hours already.
Saving the hash is meant to avoid that exactly, so stopping the analysis to save the hash should be a non-issue. If after saving the hash, and loading it, the engine is doing research, you probably want to contact the author and tell him the feature isn't working as intended.
By the way, there should be an option that makes the engine retain its analysis even if you stop/start it, should be called "preserve hash" or something like that. Ticking this box should solve the problem.
Disclaimer - I'm trying to give advice of an engine I don't have, and Learning is meant to be used for interactive analysis, not "analyzing for several hours" in where saving and loading the hash would be best.
I think you're misunderstanding me. I don't mean that after saving the hash and then stopping and starting the engine again forces the engine to analyze from scratch. No. I mean it still continues from the same depth, however, consider this scenario:
In Aquarium, let's say that the engine has been analyzing for several hrs and showing that Nf4 is the 39th possible move out of 46 that it's considering at depth 26. Nf4 is also shown in blue in the analysis window so it thinks it's interesting. If I was to stop the engine, save the hash and then reload the engine later along with the saved hash, it would start from depth 26, good...BUT, it would start from the 1st possible move out of 46, instead of starting from 39.
I hope that's more clear as this is a little hard to explain. This is the same wether I'm using R4 or Houdini. So I often have to wait for a few hrs so that it reaches 46/46 and then starts the next cycle at 1/46. Once this happens I can save the hash.
And the reason why I want it to remember where I stopped analyzing from is because I analyze for several hrs, move ahead one ply and analyze again. If it doesn't consider any new moves then the depth won't start from scratch so I move again another ply until the depth is much lower again. I then let it analyze again for a few hrs and if the new line scores better than the last time I analyzed for a few hrs then that's where I save the hash. I then repeat this process in reverse, going back one ply at a time and give the engine to consider a new move now that it knows what's coming ahead.
In Aquarium, let's say that the engine has been analyzing for several hrs and showing that Nf4 is the 39th possible move out of 46 that it's considering at depth 26. Nf4 is also shown in blue in the analysis window so it thinks it's interesting. If I was to stop the engine, save the hash and then reload the engine later along with the saved hash, it would start from depth 26, good...BUT, it would start from the 1st possible move out of 46, instead of starting from 39.
I hope that's more clear as this is a little hard to explain. This is the same wether I'm using R4 or Houdini. So I often have to wait for a few hrs so that it reaches 46/46 and then starts the next cycle at 1/46. Once this happens I can save the hash.
And the reason why I want it to remember where I stopped analyzing from is because I analyze for several hrs, move ahead one ply and analyze again. If it doesn't consider any new moves then the depth won't start from scratch so I move again another ply until the depth is much lower again. I then let it analyze again for a few hrs and if the new line scores better than the last time I analyzed for a few hrs then that's where I save the hash. I then repeat this process in reverse, going back one ply at a time and give the engine to consider a new move now that it knows what's coming ahead.
> In Aquarium, let's say that the engine has been analyzing for several hrs and showing that Nf4 is the 39th possible move out of 46 that it's considering at depth 26. Nf4 is also shown in blue in the analysis window so it thinks it's interesting. If I was to stop the engine, save the hash and then reload the engine later along with the saved hash, it would start from depth 26, good...BUT, it would start from the 1st possible move out of 46, instead of starting from 39.
Hmm... The analysis results from the first 38 moves at depth 26 are written to the hash, and should be reused after the restart.
What can change is the order of the moves to analyze - by stopping and restarting the move ordering at the root node will probably be different. This means that move numbers can be confusing after the restart, but in any event all the previous analysis results written to the hash should be reused.
It doesn't seem to work that way but I can't say you're wrong since I just can't tell for sure.
The other thing I've always wondered is what is the most efficient way of finding the best move with interactive play and perhaps long analysis both mixed in?
The way I've been doing it is analyze for a couple of hrs or so, save the line and then move up the line one ply at a time (often spending an hr or so on each ply). After about 10 plies or so depending on the position am I gaining much by back tracking one ply at a time until returning to the original position?
The other thing I've always wondered is what is the most efficient way of finding the best move with interactive play and perhaps long analysis both mixed in?
The way I've been doing it is analyze for a couple of hrs or so, save the line and then move up the line one ply at a time (often spending an hr or so on each ply). After about 10 plies or so depending on the position am I gaining much by back tracking one ply at a time until returning to the original position?
![[gb]](/mwf/v12/flags/gb.png "United Kingdom")
Both engines would reach the same depth for this test.(yorkman)
i take it you mean depths displayed.Rybkas true depth is hidden but generallly accepted to be display +3.
i take it you mean depths displayed.Rybkas true depth is hidden but generallly accepted to be display +3.
Yes, that's what I meant.
![[se]](/mwf/v12/flags/se.png "Sweden")
"I would expect Rybka 5 will surpass Houdini. Perhaps all that's needed to do this is to duplicate the changes Houdart made to create Houdini 1.5 from Ivanhoe"
Tactical search being improved <dramatically> in the next Houdini will make difference in elo anyway since it already has a stable enough overall evaluation.
This is possible to do since we already saw it in Houdini 1.5 which doesn't have better evaluation than R4 only much faster search when it gets complicated.
Why should the previous winning concept fail the next time?.
Tactical search being improved <dramatically> in the next Houdini will make difference in elo anyway since it already has a stable enough overall evaluation.
This is possible to do since we already saw it in Houdini 1.5 which doesn't have better evaluation than R4 only much faster search when it gets complicated.
Why should the previous winning concept fail the next time?.
- N/- By Banned for Life (Gold) Date 2012-01-08 20:34
If this year was a guide, Komodo MP will be like a giant among pigmies in next years tournament...
Experts agree!
Response-
If this year was a guide, Komodo MP will be like a giant among pigmies in next years tournament...
Experts agree!
Response-
![[ru]](/mwf/v12/flags/ru.png "Russian Federation")
Sorry for this large pic, but it looks delicious
What's wrong with the guy at the very right? He never dares to actually lick the fish.
> What's wrong with the guy at the very right? He never dares to actually lick the fish.
Maybe he doesn't like tail????
Just not when it smells fishy
I'm in the weird part of Rybka forum again :)
Yes that is what it looks like
I just hope it doesn't turn out to be a myth like the Loch Ness Monster :-)
Even myths have some basis in reality!
Don't forget to zoom out the picture!
That pussy is going to lose its face!
You have watched too many cartoons!
And you haven't watched enough!
A pussy- isn't even representative of Critter!
You've lead a sheltered existences- or, have never been to the flicks. And-NO! It's not the chick!
A pussy- isn't even representative of Critter!
You've lead a sheltered existences- or, have never been to the flicks. And-NO! It's not the chick!
> A pussy- isn't even representative of Critter!
It is if it's going to eat all those fishes
LOL, what an awesome graphic.
I seriously wouldn't stick my face in that fish bowl if I were kitty though.
I seriously wouldn't stick my face in that fish bowl if I were kitty though.
Rybka executable was too large. I wish it will be much more compact like this file
I never found the executables too large, for instance, all 73 freeware Rybkas fit in a small 4.36MB file.
No, no, no, no!
Still could be smaller like 3.638 MB in same 73 files. No limit to perfectness!
But thanks :)
Still could be smaller like 3.638 MB in same 73 files. No limit to perfectness!
But thanks :)
Attachment: RybkaFree.nz (3726k)
....nz? What's that?
NanoZip, can often compress faster than RAR/7-Zip and smaller if needed, at PAQ/cm lvl.
Experimental still...
Imagine it's the year 1994... Rybka? what is that?
Experimental still...
Imagine it's the year 1994... Rybka? what is that?
> Imagine it's the year 1994... Rybka? what is that?
...Or maybe year 2000? I find the compression impressive but I wonder if technology of year 2023 can get the file down to 200Kb
Anyway, thanks for the info, I have to check it out.
how do I recompile crafty to 16 bits from source code so it works with windows 3.21?
Once Rybka 5 is out, and Vas shows the world Rybka is stronger than Houdini 3 (it is very possible, anything is Possible), the engine room in playchess will be full of Rybkas like in the old days. Hopefully it will make a strong comeback, I can't wait to buy the new Rybka engine. All the good vibrations and good energy from me and all of us Rybka enthusiasts to your Rybka 5 and you Mr. Vas.
Assuming that there are inherent difficulties in programming increasing elo. Considering that Rybka already is a fine analysis tool- Vas wouldn't be remiss in going the way of- say, Richard Vida and leaving the challenge of elo for a more superior functioning chess engine analysis tool. My guess would be that Rybka 5 would regardless still end up being a highly competitive UCI chess engine.
Addendum:
I think the majority would favor a superior analysis tool over strength at point.
Addendum:
I think the majority would favor a superior analysis tool over strength at point.
Probably right, but a little extra strength would not hurt
Considering Robert Houdart's latest declaration re: elo. I have know idea where Vas is in his development of Rybka 5 .
However, he did say in so many words that it would be a serious analysis tool.
That aside, and because Rybka 4 wasn't- for the first time in so many years, the number 1 chess engine-Vas' credibility for his willingness to go the extra mile has come into question.
So, knowing the elo gap he may now need to cover to regain the lead - how much elo strength concomitant with a superior analysis tool would be acceptable for most users in supporting Vas as a major participant in UCI chess engine development.
I guess the question comes down to- have the priorities in chess engine development changed or are they changing? If so, then to what degree? Which leads back to your statement.
However, he did say in so many words that it would be a serious analysis tool.
That aside, and because Rybka 4 wasn't- for the first time in so many years, the number 1 chess engine-Vas' credibility for his willingness to go the extra mile has come into question.
So, knowing the elo gap he may now need to cover to regain the lead - how much elo strength concomitant with a superior analysis tool would be acceptable for most users in supporting Vas as a major participant in UCI chess engine development.
I guess the question comes down to- have the priorities in chess engine development changed or are they changing? If so, then to what degree? Which leads back to your statement.
> I guess the question comes down to- have the priorities in chess engine development changed or are they changing?
I think they definitively are changing. Also I think all chess programmers should make their programs into something that can analyze as well as play. The playing field is getting smaller.
Yeah, and all GUI's work as intended!
It looks like by the time I build my new hardware, Rybka 5 will be out and ready to kick some butt, and show to the world Rybka 5 is the one and only King of all engines.
1 more day scratch off my calendar.
1 more day scratch off my calendar.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill