Trying out the Leela Hybrid engine (NN with AB)

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
corres
Posts: 1558
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by corres » Sun Mar 03, 2019 9:33 am

Ozymandias wrote:
Sun Mar 03, 2019 9:15 am
pohl4711 wrote:
Sun Mar 03, 2019 8:02 am
A fixed movetime gives much higher search depths in endgame and does not depend on the chosen engine, the used hardware and the number of threads.
Where do you get that "go movetime N" is independent of engine and HW? A faster engine will reach higher depth with the same allocated time, similarly, faster HW or a higher thread count will have the same effect; you must know that.
Please, read once again my post about my test.
Ozymandias wrote:
Sun Mar 03, 2019 9:15 am
I for one, would prefer for both options to be kept on the table, let the user decide if they want to go with fixed depth or not.
There is no any obstacle to modify the source of Leelafish to fixed "depth" and fixed "movetime" being together and users can choose the method what they want.
As an earlier post I mentioned too much "help" for Leela may decrease the Elo of Leela.
It was confirmed by dkappe too.

corres
Posts: 1558
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by corres » Sun Mar 03, 2019 9:36 am

Ozymandias wrote:
Sun Mar 03, 2019 9:21 am
corres wrote:
Sun Mar 03, 2019 8:40 am
The large majority of games are decided in the middle game and not in the endgame.
That would be in line with what I recall from Laskos' test, you won't happen to remember which thread that was, will you?
Without tests of Laskos it is an old experience.

User avatar
Ozymandias
Posts: 1102
Joined: Sun Oct 25, 2009 12:30 am

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by Ozymandias » Sun Mar 03, 2019 9:53 am

corres wrote:
Sun Mar 03, 2019 9:33 am
Ozymandias wrote:
Sun Mar 03, 2019 9:15 am
Where do you get that "go movetime N" is independent of engine and HW? A faster engine will reach higher depth with the same allocated time, similarly, faster HW or a higher thread count will have the same effect; you must know that.
Please, read once again my post about my test.
Can you post a game with depth/time info?

corres
Posts: 1558
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by corres » Sun Mar 03, 2019 11:51 am

Ozymandias wrote:
Sun Mar 03, 2019 9:53 am
corres wrote:
Sun Mar 03, 2019 9:33 am
Ozymandias wrote:
Sun Mar 03, 2019 9:15 am
Where do you get that "go movetime N" is independent of engine and HW? A faster engine will reach higher depth with the same allocated time, similarly, faster HW or a higher thread count will have the same effect; you must know that.
Please, read once again my post about my test.
Can you post a game with depth/time info?
http://wikisend.com/download/716842/leelafish-test.zip

Password:
leelafish

User avatar
Ozymandias
Posts: 1102
Joined: Sun Oct 25, 2009 12:30 am

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by Ozymandias » Sun Mar 03, 2019 12:37 pm

corres wrote:
Sun Mar 03, 2019 11:51 am
Ozymandias wrote:
Sun Mar 03, 2019 9:53 am
corres wrote:
Sun Mar 03, 2019 9:33 am
Ozymandias wrote:
Sun Mar 03, 2019 9:15 am
Where do you get that "go movetime N" is independent of engine and HW? A faster engine will reach higher depth with the same allocated time, similarly, faster HW or a higher thread count will have the same effect; you must know that.
Please, read once again my post about my test.
Can you post a game with depth/time info?
http://wikisend.com/download/716842/leelafish-test.zip

Password:
leelafish
We are sorry, but the requested file doesn't exist or has expired and is no longer available.

corres
Posts: 1558
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by corres » Sun Mar 03, 2019 2:03 pm

Ozymandias wrote:
Sun Mar 03, 2019 12:37 pm
corres wrote:
Sun Mar 03, 2019 11:51 am
Ozymandias wrote:
Sun Mar 03, 2019 9:53 am
corres wrote:
Sun Mar 03, 2019 9:33 am
Ozymandias wrote:
Sun Mar 03, 2019 9:15 am
Where do you get that "go movetime N" is independent of engine and HW? A faster engine will reach higher depth with the same allocated time, similarly, faster HW or a higher thread count will have the same effect; you must know that.
Please, read once again my post about my test.
Can you post a game with depth/time info?
http://wikisend.com/download/716842/leelafish-test.zip

Password:
leelafish
We are sorry, but the requested file doesn't exist or has expired and is no longer available.
I am afraid you can not use this site.
After this message click on "download page" and type in the window File Id:
716842
and click "Download file" and type the password, etc.

User avatar
Ozymandias
Posts: 1102
Joined: Sun Oct 25, 2009 12:30 am

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by Ozymandias » Sun Mar 03, 2019 4:25 pm

corres wrote:
Sun Mar 03, 2019 2:03 pm
Ozymandias wrote:
Sun Mar 03, 2019 12:37 pm
corres wrote:
Sun Mar 03, 2019 11:51 am
Ozymandias wrote:
Sun Mar 03, 2019 9:53 am
corres wrote:
Sun Mar 03, 2019 9:33 am
Ozymandias wrote:
Sun Mar 03, 2019 9:15 am
Where do you get that "go movetime N" is independent of engine and HW? A faster engine will reach higher depth with the same allocated time, similarly, faster HW or a higher thread count will have the same effect; you must know that.
Please, read once again my post about my test.
Can you post a game with depth/time info?
http://wikisend.com/download/716842/leelafish-test.zip

Password:
leelafish
We are sorry, but the requested file doesn't exist or has expired and is no longer available.
I am afraid you can not use this site.
After this message click on "download page" and type in the window File Id:
716842
and click "Download file" and type the password, etc.
As I suspected, I can see Lc0's depths, but not the auxiliary engine's, which means no comparison can be made, in terms of depth reached with a fixed time per move TC.

corres
Posts: 1558
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by corres » Sun Mar 03, 2019 5:09 pm

Ozymandias wrote:
Sun Mar 03, 2019 4:25 pm

As I suspected, I can see Lc0's depths, but not the auxiliary engine's, which means no comparison can be made, in terms of depth reached with a fixed time per move TC.
This is not the part of Leela's pv.
The reached depth obviously depends on the value of the fixed movetime, the speed of PC and the actual number of moves at that position.
As I noted on my PC 100 ms movetime is about the average time to reach depth = 15.

kalikoba2019
Posts: 8
Joined: Sun Mar 03, 2019 11:46 am
Full name: Kalikoba

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by kalikoba2019 » Tue Mar 05, 2019 10:07 am

Hello Killerducky, I have a special request for engine parameters : could you please modify the engine so that it can accept to use .bin opening book please ?
I know that it is possible from Stockfish to do it, like we have brainfish having those parameters.
This is because i make engine matches on the internet and unfortunately, using polyglot won't work.
Needed parameters would be :
Bookfile= <path of the engine>
Bestbookmove=<true or false, depending on if we want the best move to always be picked>
BookRandom=<true or false, depending on if we want best book move or not
BookDepth=<number of semiplies we want the engine to use the book>

Thank you,

Best regards

brianr
Posts: 356
Joined: Thu Mar 09, 2006 2:01 pm

Re: Trying out the Leela Hybrid engine (NN with AB)

Post by brianr » Tue Mar 05, 2019 12:13 pm

Just wondering why Polyglot with a book does not work for you.
Thanks

Some parameters:
http://manpages.ubuntu.com/manpages/bio ... lot.6.html

Post Reply