Feature request from Robert Houdart

Discussion about chess-playing software (engines, hosts, opening books, platforms, etc...)
LucenaTheLucid
Posts: 160
Joined: Thu Jun 10, 2010 2:14 am
Real Name: Luis Smith

Feature request from Robert Houdart

Post by LucenaTheLucid » Sun Jul 11, 2010 1:04 am

Hi Robert I hope you don't mind if I ask that you implement some sort of tuning parameters in your next version of Houdini. Piece values, as well as some sort of aggressiveness settings would be fantastic! From the stand point of a user, I think this engine gives the best analysis out of all the engines on the market, free or otherwise. So thanks again Robert for a great engine.

User avatar
Robert Houdart
Posts: 180
Joined: Thu Jun 10, 2010 4:55 pm
Contact:

Re: Feature request from Robert Houdart

Post by Robert Houdart » Sun Jul 11, 2010 10:47 am

Luis, after Houdini 1.03 I will be looking at adding some options (as few as possible) to control the aggressiveness of the engine.

I'm slightly reluctant to add zillions of UCI options for individual piece values, time control parameters or extensions, in my opinion it's the responsibility of the engine author to provide the best possible values for these settings.
But I'm open to any reasonable suggestion, if many users want to play with these parameters I may reconsider.

Robert

BTO7
Posts: 101
Joined: Thu Jun 10, 2010 4:21 am

Re: Feature request from Robert Houdart

Post by BTO7 » Sun Jul 11, 2010 1:03 pm

Robert i think you hit the nail on the head. Keep it simple. I'm all for as few as possible although keep in mind with tuning suits such as STS its nice to fine tune a piece to your specific machine and operating system. The question becomes is it worth the extra coding and possibly some slow down because of having the option? Again keep it simple. I'm sure you will make a great choice on how best to balance these issues and keep up the great work. Houdini is light years ahead of the rest in the speed category I personally would hate to sac that for too many options myself ;)

Regards
BT

User avatar
Matthias Gemuh
Posts: 295
Joined: Wed Jun 09, 2010 2:48 pm
Contact:

Re: Feature request from Robert Houdart

Post by Matthias Gemuh » Sun Jul 11, 2010 2:39 pm

BTO7 wrote:Robert i think you hit the nail on the head. Keep it simple. I'm all for as few as possible although keep in mind with tuning suits such as STS its nice to fine tune a piece to your specific machine and operating system. The question becomes is it worth the extra coding and possibly some slow down because of having the option? Again keep it simple. I'm sure you will make a great choice on how best to balance these issues and keep up the great work. Houdini is light years ahead of the rest in the speed category I personally would hate to sac that for too many options myself ;)

Regards
BT
I agree. Fire(bird) and Ivanhoe have too many options that should never have been offered to end users.

Matthias.
Aided by engines, GMs can be very strong.
http://www.hylogic.de

LucenaTheLucid
Posts: 160
Joined: Thu Jun 10, 2010 2:14 am
Real Name: Luis Smith

Re: Feature request from Robert Houdart

Post by LucenaTheLucid » Sun Jul 11, 2010 3:28 pm

Matthias Gemuh wrote:
BTO7 wrote:Robert i think you hit the nail on the head. Keep it simple. I'm all for as few as possible although keep in mind with tuning suits such as STS its nice to fine tune a piece to your specific machine and operating system. The question becomes is it worth the extra coding and possibly some slow down because of having the option? Again keep it simple. I'm sure you will make a great choice on how best to balance these issues and keep up the great work. Houdini is light years ahead of the rest in the speed category I personally would hate to sac that for too many options myself ;)

Regards
BT
I agree. Fire(bird) and Ivanhoe have too many options that should never have been offered to end users.

Matthias.
Hi Matthias,

While I agree Firebird, IvanHoe and others have way too many options many of which I do not understand, I very much loved the tunable parameters in BigLion, Stockfish, and even TheKing 3.50. I was even thinking of making a series of 'personalities' for Stockfish such as old World Chess Champions and other Grandmasters.

I think as long as the parameters are kept to a minimum (10-15) and also easy to understand for users without chess programming knowledge (i.e. Pawn Value: 100, Bishop Pair: 100 etc...) it would be great. Hopefully though it will not affect the program too much.

User avatar
Dr.Wael Deeb
Posts: 104
Joined: Thu Jun 10, 2010 8:29 pm
Real Name: Dr.Wael Deeb

Re: Feature request from Robert Houdart

Post by Dr.Wael Deeb » Sun Jul 11, 2010 4:03 pm

LucenaTheLucid wrote:
Matthias Gemuh wrote:
BTO7 wrote:Robert i think you hit the nail on the head. Keep it simple. I'm all for as few as possible although keep in mind with tuning suits such as STS its nice to fine tune a piece to your specific machine and operating system. The question becomes is it worth the extra coding and possibly some slow down because of having the option? Again keep it simple. I'm sure you will make a great choice on how best to balance these issues and keep up the great work. Houdini is light years ahead of the rest in the speed category I personally would hate to sac that for too many options myself ;)

Regards
BT
I agree. Fire(bird) and Ivanhoe have too many options that should never have been offered to end users.

Matthias.
Hi Matthias,

While I agree Firebird, IvanHoe and others have way too many options many of which I do not understand, I very much loved the tunable parameters in BigLion, Stockfish, and even TheKing 3.50. I was even thinking of making a series of 'personalities' for Stockfish such as old World Chess Champions and other Grandmasters.

I think as long as the parameters are kept to a minimum (10-15) and also easy to understand for users without chess programming knowledge (i.e. Pawn Value: 100, Bishop Pair: 100 etc...) it would be great. Hopefully though it will not affect the program too much.
I have already a FireBird 1.1 personality which plays equal to Deep Rybka 4,but I have to cross the 400 games barrier before I announce it....FireBird 1.1 EP8+ which I renamed to WD+ by the way,is already stronger than Rybka 3 MP and all of the Ippo family monsters....

Yet,I didn't test Fire 1.31,IvanHoe T55 and Houdini,so we'll see :D
Cheers,
Dr.D

User avatar
Swaminathan
Posts: 375
Joined: Wed Jun 09, 2010 12:14 pm

Re: Feature request from Robert Houdart

Post by Swaminathan » Sun Jul 11, 2010 4:11 pm

+1

I also love engine with option to change values for certain pieces, mobility, bonus, passedpawn, aggressiveness, bishop pair, kingsafety etc . Preferably likes the ones in Chessmaster. :)
tweaking and optimising the settings to try and gain more elo in what's already a 3300 engine would be awesome.
Logo made by Ulysses P (Vytron)
Co-Authored with Dann Corbit: Strategic Test Suite

look
Posts: 6
Joined: Sat Jun 12, 2010 8:15 am
Real Name: Mehdi Amini Valashani son of Bahram
Location: Iran

Re: Feature request from Robert Houdart

Post by look » Sun Jul 11, 2010 4:38 pm

Robert Houdart wrote:Luis, after Houdini 1.03 I will be looking at adding some options (as few as possible) to control the aggressiveness of the engine.

I'm slightly reluctant to add zillions of UCI options for individual piece values, time control parameters or extensions, in my opinion it's the responsibility of the engine author to provide the best possible values for these settings.
But I'm open to any reasonable suggestion, if many users want to play with these parameters I may reconsider.

Robert
Hi,

i suggest a few more parameters too, but only those that either potentially improve playing strength or give a different personality.

User avatar
Matthias Gemuh
Posts: 295
Joined: Wed Jun 09, 2010 2:48 pm
Contact:

Re: Feature request from Robert Houdart

Post by Matthias Gemuh » Sun Jul 11, 2010 4:52 pm

LucenaTheLucid wrote:
Matthias Gemuh wrote:
... Fire(bird) and Ivanhoe have too many options that should never have been offered to end users.

Matthias.
Hi Matthias,

While I agree Firebird, IvanHoe and others have way too many options many of which I do not understand, I very much loved the tunable parameters in BigLion, Stockfish, and even TheKing 3.50. I was even thinking of making a series of 'personalities' for Stockfish such as old World Chess Champions and other Grandmasters.

I think as long as the parameters are kept to a minimum (10-15) and also easy to understand for users without chess programming knowledge (i.e. Pawn Value: 100, Bishop Pair: 100 etc...) it would be great. Hopefully though it will not affect the program too much.

Yes, there exists an optimal number of tunable parameters that should be externalized.
Everything else should remain in the black box. Opening it only makes it a Pandora's box.

Matthias.
Aided by engines, GMs can be very strong.
http://www.hylogic.de

BTO7
Posts: 101
Joined: Thu Jun 10, 2010 4:21 am

Re: Feature request from Robert Houdart

Post by BTO7 » Sun Jul 11, 2010 5:59 pm

Ok food for thought. The best with less in my opinion and why. Short story long. I have found thanks to STS some interesting things. First one must find the positions that the engine misses and then go back over that and figure out why. Here is what i have found thats interesting. In the case of the bishops.....I had a epd position in which the engine had to choose between both bishops as the number 1 and 2 moves. Now no amount of adjusting the piece values i found to help the engine to decide what bishop was the better move. This finally led me to a adjustment (in this case PP Ivan55mPP) that did help the engine decide and it was not the piece value at all ! Turned out why the engine was confused on what bishop to play was because the one move had the black bishop coming back to protect the king a bit more while the best move was the other bishop coming to a more aggressive file. When i turned down KING SAFETY suddenly it found the move faster and sooner. This puts KING SAFETY on my list. In another position the engine was having a hard time choosing between the Q and bishop moves and in this case lowering material weighting value helped the engine find the best move. This engine does not have positional tuning and now i wish it did. In the end what i have found to really help is .....Knights being able to be tuned......King safety......material weighting....and positional weighting along with the bishop pair the most helpful to the tuner. If the programmer ...you in this case Robert :) can do the rest internally . I think these few options will give us tuners enough to work with ...with out weighting down things. You find the best queen, bishop, rook and pawn values and we can do the rest.

Post Reply