|

»

Gnu - Backgammon

Feature request: Training tool "Randomize first N moves within 0.XXX equity."

View: New views

8 Messages — Rating Filter: Alert me

	Feature request: Training tool "Randomize first N moves within 0.XXX equity." by Michael Depreli :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message Some parts of this message have been removed. Learn more about Nabble's security policy. I'm sure many users use gnubg simply to play against as self training. One issue that for me doesn't help with training is that gnubg will obviously always play the same opening move or reply etc in the same given position. In real life backgammon obviously this isn't the case. As a simple example for money gnubg plays 52 13/8 13/11. Rollouts show this to be an error, and I'm guessing most humans would not play this move the majority of the time. Also rollouts have shown many opening plays and replies are very close in equity so as not be be considered an error. My request therefore is to have an option to randomize the first N moves within equity X. I would suggest something like a move filter: E.g. Randomize first N moves within X plus additional N moves within Y. Hopefully I'm not the only user that would find this feature useful. Michael Have more than one Hotmail account? Link them together to easily access both. _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg
	Re: Feature request: Training tool "Randomize first N moves within 0.XXX equity." by Jonathan Kinsey :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message Some parts of this message have been removed. Learn more about Nabble's security policy. This reminds me of the discussion on "personalities", as the kind of approach you suggest could easily fit into the post-evaluation equity adjustments that are an easy way to add varied gnu-opponents. Jon Michael Depreli wrote: > I'm sure many users use gnubg simply to play against as self training. > One issue that for me doesn't help with training is that gnubg will > obviously always play the same opening move or reply etc in the same > given position. > In real life backgammon obviously this isn't the case. > As a simple example for money gnubg plays 52 13/8 13/11. > Rollouts show this to be an error, and I'm guessing most humans would > not play this move the majority of the time. > Also rollouts have shown many opening plays and replies are very close > in equity so as not be be considered an error. > > My request therefore is to have an option to randomize the first N moves > within equity X. > > I would suggest something like a move filter: > > E.g. Randomize first N moves within X plus additional N moves within Y. > > Hopefully I'm not the only user that would find this feature useful. > > > Michael > > > > > > ------------------------------------------------------------------------ > Have more than one Hotmail account? Link them together to easily access > both. > > > ------------------------------------------------------------------------ > > _______________________________________________ > Bug-gnubg mailing list > Bug-gnubg@... > http://lists.gnu.org/mailman/listinfo/bug-gnubg New! Receive and respond to mail from other email accounts from within Hotmail Find out how. _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg
	Re: Feature request: Training tool "Randomize first N moves within 0.XXX equity." by Massimiliano Maini-3 :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message The "personalities" thing would be very interesting IMO: - a small set of predefined personalities (eventually, ability to define custom ones, on chequer play and cube decision) - possibility to chose a given personality or to pick a random one (not disclosed, unless requested). MaX. 2009/10/8 Jonathan Kinsey <jon_kinsey@...>: > This reminds me of the discussion on "personalities", as the kind of > approach > you suggest could easily fit into the post-evaluation equity adjustments > that > are an easy way to add varied gnu-opponents. > > Jon > > Michael Depreli wrote: >> I'm sure many users use gnubg simply to play against as self training. >> One issue that for me doesn't help with training is that gnubg will >> obviously always play the same opening move or reply etc in the same >> given position. >> In real life backgammon obviously this isn't the case. >> As a simple example for money gnubg plays 52 13/8 13/11. >> Rollouts show this to be an error, and I'm guessing most humans would >> not play this move the majority of the time. >> Also rollouts have shown many opening plays and replies are very close >> in equity so as not be be considered an error. >> >> My request therefore is to have an option to randomize the first N moves >> within equity X. >> >> I would suggest something like a move filter: >> >> E.g. Randomize first N moves within X plus additional N moves within Y. >> >> Hopefully I'm not the only user that would find this feature useful. >> >> >> Michael >> >> >> >> >> >> ------------------------------------------------------------------------ >> Have more than one Hotmail account? Link them together to easily access >> both. >> >> >> ------------------------------------------------------------------------ >> >> _______________________________________________ >> Bug-gnubg mailing list >> Bug-gnubg@... >> http://lists.gnu.org/mailman/listinfo/bug-gnubg > > > > > ________________________________ > New! Receive and respond to mail from other email accounts from within > Hotmail Find out how. > _______________________________________________ > Bug-gnubg mailing list > Bug-gnubg@... > http://lists.gnu.org/mailman/listinfo/bug-gnubg > > _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg
	RE: Feature request: Training tool "Randomize first N moves within 0.XXX equity." by Michael Depreli :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message Some parts of this message have been removed. Learn more about Nabble's security policy. You could have "aggressive personality" ! Defined as maybe if moves are within x equity chose the best one that hits a checker. Also for cube decisions if volatility >X then double earlier within X equity. Lots of possibilities but I'd be happy just being able to randomize the openings and replies. That alone would would lead to many different games for practicing. Michael > From: maxmaini@... > Date: Thu, 8 Oct 2009 15:59:06 +0200 > Subject: Re: [Bug-gnubg] Feature request: Training tool "Randomize first N moves within 0.XXX equity." > To: jon_kinsey@... > CC: michaeldepreli@...; bug-gnubg@... > > The "personalities" thing would be very interesting IMO: > > - a small set of predefined personalities (eventually, ability to > define custom ones, on chequer play and cube decision) > > - possibility to chose a given personality or to pick a random > one (not disclosed, unless requested). > > MaX. > > 2009/10/8 Jonathan Kinsey <jon_kinsey@...>: > > This reminds me of the discussion on "personalities", as the kind of > > approach > > you suggest could easily fit into the post-evaluation equity adjustments > > that > > are an easy way to add varied gnu-opponents. > > > > Jon > > > > Michael Depreli wrote: > >> I'm sure many users use gnubg simply to play against as self training. > >> One issue that for me doesn't help with training is that gnubg will > >> obviously always play the same opening move or reply etc in the same > >> given position. > >> In real life backgammon obviously this isn't the case. > >> As a simple example for money gnubg plays 52 13/8 13/11. > >> Rollouts show this to be an error, and I'm guessing most humans would > >> not play this move the majority of the time. > >> Also rollouts have shown many opening plays and replies are very close > >> in equity so as not be be considered an error. > >> > >> My request therefore is to have an option to randomize the first N moves > >> within equity X. > >> > >> I would suggest something like a move filter: > >> > >> E.g. Randomize first N moves within X plus additional N moves within Y. > >> > >> Hopefully I'm not the only user that would find this feature useful. > >> > >> > >> Michael > >> > >> > >> > >> > >> > >> ------------------------------------------------------------------------ > >> Have more than one Hotmail account? Link them together to easily access > >> both. > >> > >> > >> ------------------------------------------------------------------------ > >> > >> _______________________________________________ > >> Bug-gnubg mailing list > >> Bug-gnubg@... > >> http://lists.gnu.org/mailman/listinfo/bug-gnubg > > > > > > > > > > ________________________________ > > New! Receive and respond to mail from other email accounts from within > > Hotmail Find out how. > > _______________________________________________ > > Bug-gnubg mailing list > > Bug-gnubg@... > > http://lists.gnu.org/mailman/listinfo/bug-gnubg > > > > New! Receive and respond to mail from other email accounts from within Hotmail Find out how. _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg
	RE: Feature request: Training tool "Randomize first N moves within 0.XXX equity." by Michael Depreli :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message Some parts of this message have been removed. Learn more about Nabble's security policy. Also for "Aggressive" if moves are within x equity chose the best one that hits a checker OR one that leaves a blot. From: michaeldepreli@... To: maxmaini@...; jon_kinsey@... Subject: RE: [Bug-gnubg] Feature request: Training tool "Randomize first N moves within 0.XXX equity." Date: Thu, 8 Oct 2009 14:46:03 +0000 CC: bug-gnubg@... You could have "aggressive personality" ! Defined as maybe if moves are within x equity chose the best one that hits a checker. Also for cube decisions if volatility >X then double earlier within X equity. Lots of possibilities but I'd be happy just being able to randomize the openings and replies. That alone would would lead to many different games for practicing. Michael > From: maxmaini@... > Date: Thu, 8 Oct 2009 15:59:06 +0200 > Subject: Re: [Bug-gnubg] Feature request: Training tool "Randomize first N moves within 0.XXX equity." > To: jon_kinsey@... > CC: michaeldepreli@...; bug-gnubg@... > > The "personalities" thing would be very interesting IMO: > > - a small set of predefined personalities (eventually, ability to > define custom ones, on chequer play and cube decision) > > - possibility to chose a given personality or to pick a random > one (not disclosed, unless requested). > > MaX. > > 2009/10/8 Jonathan Kinsey <jon_kinsey@...>: > > This reminds me of the discussion on "personalities", as the kind of > > approach > > you suggest could easily fit into the post-evaluation equity adjustments > > that > > are an easy way to add varied gnu-opponents. > > > > Jon > > > > Michael Depreli wrote: > >> I'm sure many users use gnubg simply to play against as self training. > >> One issue that for me doesn't help with training is that gnubg will > >> obviously always play the same opening move or reply etc in the same > >> given position. > >> In real life backgammon obviously this isn't the case. > >> As a simple example for money gnubg plays 52 13/8 13/11. > >> Rollouts show this to be an error, and I'm guessing most humans would > >> not play this move the majority of the time. > >> Also rollouts have shown many opening plays and replies are very close > >> in equity so as not be be considered an error. > >> > >> My request therefore is to have an option to randomize the first N moves > >> within equity X. > >> > >> I would suggest something like a move filter: > >> > >> E.g. Randomize first N moves within X plus additional N moves within Y. > >> > >> Hopefully I'm not the only user that would find this feature useful. > >> > >> > >> Michael > >> > >> > >> > >> > >> > >> ------------------------------------------------------------------------ > >> Have more than one Hotmail account? Link them together to easily access > >> both. > >> > >> > >> ------------------------------------------------------------------------ > >> > >> _______________________________________________ > >> Bug-gnubg mailing list > >> Bug-gnubg@... > >> http://lists.gnu.org/mailman/listinfo/bug-gnubg > > > > > > > > > > ________________________________ > > New! Receive and respond to mail from other email accounts from within > > Hotmail Find out how. > > _______________________________________________ > > Bug-gnubg mailing list > > Bug-gnubg@... > > http://lists.gnu.org/mailman/listinfo/bug-gnubg > > > > New! Receive and respond to mail from other email accounts from within Hotmail Find out how. Use Hotmail to send and receive mail from your different email accounts. Find out how. _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg
	Re: Feature request: Training tool "Randomize first N moves within 0.XXX equity." by Jonathan Kinsey :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message Some parts of this message have been removed. Learn more about Nabble's security policy. Yes, this was all discussed a while ago, there's a few obvious things that could be applied (like blots, take points etc). Like all these things it's a matter of time, realistically things like this are post v1 (unless someone decides to do them...). Jon Michael Depreli wrote: > You could have "aggressive personality" ! > Defined as maybe if moves are within x equity chose the best one that > hits a checker. > Also for cube decisions if volatility>X then double earlier within X > equity. > > Lots of possibilities but I'd be happy just being able to randomize the > openings and replies. > That alone would would lead to many different games for practicing. > > > Michael > > > >> From: maxmaini@... >> Date: Thu, 8 Oct 2009 15:59:06 +0200 >> Subject: Re: [Bug-gnubg] Feature request: Training tool "Randomize > first N moves within 0.XXX equity." >> To: jon_kinsey@... >> CC: michaeldepreli@...; bug-gnubg@... >> >> The "personalities" thing would be very interesting IMO: >> >> - a small set of predefined personalities (eventually, ability to >> define custom ones, on chequer play and cube decision) >> >> - possibility to chose a given personality or to pick a random >> one (not disclosed, unless requested). >> >> MaX. >> >> 2009/10/8 Jonathan Kinsey : >>> This reminds me of the discussion on "personalities", as the kind of >>> approach >>> you suggest could easily fit into the post-evaluation equity adjustments >>> that >>> are an easy way to add varied gnu-opponents. >>> >>> Jon >>> >>> Michael Depreli wrote: >>>> I'm sure many users use gnubg simply to play against as self training. >>>> One issue that for me doesn't help with training is that gnubg will >>>> obviously always play the same opening move or reply etc in the same >>>> given position. >>>> In real life backgammon obviously this isn't the case. >>>> As a simple example for money gnubg plays 52 13/8 13/11. >>>> Rollouts show this to be an error, and I'm guessing most humans would >>>> not play this move the majority of the time. >>>> Also rollouts have shown many opening plays and replies are very close >>>> in equity so as not be be considered an error. >>>> >>>> My request therefore is to have an option to randomize the first N > moves >>>> within equity X. >>>> >>>> I would suggest something like a move filter: >>>> >>>> E.g. Randomize first N moves within X plus additional N moves within Y. >>>> >>>> Hopefully I'm not the only user that would find this feature useful. >>>> >>>> >>>> Michael >>>> >>>> >>>> >>>> >>>> >>>> > ------------------------------------------------------------------------ >>>> Have more than one Hotmail account? Link them together to easily access >>>> both. >>>> >>>> >>>> > ------------------------------------------------------------------------ >>>> >>>> _______________________________________________ >>>> Bug-gnubg mailing list >>>> Bug-gnubg@... >>>> http://lists.gnu.org/mailman/listinfo/bug-gnubg >>> >>> >>> >>> >>> ________________________________ >>> New! Receive and respond to mail from other email accounts from within >>> Hotmail Find out how. >>> _______________________________________________ >>> Bug-gnubg mailing list >>> Bug-gnubg@... >>> http://lists.gnu.org/mailman/listinfo/bug-gnubg >>> >>> > > ------------------------------------------------------------------------ > New! Receive and respond to mail from other email accounts from within > Hotmail Find out how. View your other email accounts from your Hotmail inbox. Add them now. _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg
	Re: Feature request: Training tool "Randomize first N moves within 0.XXX equity." by Michael Petch :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message Re: [Bug-gnubg] Feature request: Training tool "Randomize first N moves within 0.XXX equity." Another variation on a theme one might be to not double on an “Optional cube” and wait until it becomes an actual Double On 08/10/09 8:46 AM, "michaeldepreli@..." <michaeldepreli@...> wrote: You could have "aggressive personality" ! Defined as maybe if moves are within x equity chose the best one that hits a checker. Also for cube decisions if volatility >X then double earlier within X equity. Lots of possibilities but I'd be happy just being able to randomize the openings and replies. That alone would would lead to many different games for practicing. Michael > From: maxmaini@... > Date: Thu, 8 Oct 2009 15:59:06 +0200 > Subject: Re: [Bug-gnubg] Feature request: Training tool "Randomize first N moves within 0.XXX equity." > To: jon_kinsey@... > CC: michaeldepreli@...; bug-gnubg@... > > The "personalities" thing would be very interesting IMO: > > - a small set of predefined personalities (eventually, ability to > define custom ones, on chequer play and cube decision) > > - possibility to chose a given personality or to pick a random > one (not disclosed, unless requested). > > MaX. > > 2009/10/8 Jonathan Kinsey <jon_kinsey@...>: > > This reminds me of the discussion on "personalities", as the kind of > > approach > > you suggest could easily fit into the post-evaluation equity adjustments > > that > > are an easy way to add varied gnu-opponents. > > > > Jon > > > > Michael Depreli wrote: > >> I'm sure many users use gnubg simply to play against as self training. > >> One issue that for me doesn't help with training is that gnubg will > >> obviously always play the same opening move or reply etc in the same > >> given position. > >> In real life backgammon obviously this isn't the case. > >> As a simple example for money gnubg plays 52 13/8 13/11. > >> Rollouts show this to be an error, and I'm guessing most humans would > >> not play this move the majority of the time. > >> Also rollouts have shown many opening plays and replies are very close > >> in equity so as not be be considered an error. > >> > >> My request therefore is to have an option to randomize the first N moves > >> within equity X. > >> > >> I would suggest something like a move filter: > >> > >> E.g. Randomize first N moves within X plus additional N moves within Y. > >> > >> Hopefully I'm not the only user that would find this feature useful. > >> > >> > >> Michael > >> > >> > >> > >> > >> > >> ------------------------------------------------------------------------ > >> Have more than one Hotmail account? Link them together to easily access > >> both. > >> > >> > >> ------------------------------------------------------------------------ > >> > >> _______________________________________________ > >> Bug-gnubg mailing list > >> Bug-gnubg@... > >> http://lists.gnu.org/mailman/listinfo/bug-gnubg > > > > > > > > > > ________________________________ > > New! Receive and respond to mail from other email accounts from within > > Hotmail Find out how. > > _______________________________________________ > > Bug-gnubg mailing list > > Bug-gnubg@... > > http://lists.gnu.org/mailman/listinfo/bug-gnubg > > > > New! Receive and respond to mail from other email accounts from within Hotmail Find out how. <http://clk.atdmt.com/UKM/go/167688463/direct/01/> _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg
	Re: Feature request: Training tool "Randomize first N moves within 0.XXX equity." by Massimiliano Maini-3 :: Rate this Message: Reply to Author \| View Threaded \| Show Only this Message For chequer play, I can see: - favor the move than wins more gammons (when gammons count, maybe) and/or (maybe the two overlap): - favor/disfavor the move that hits (or the move that blots) For cube action: - cube aggressive/cube shy (double close no-doubles and take close passes / no-double close doubles, pass close takes) For each "presonality aspect", multiple levels could be defined, e.g.: - gammon-go attitude: disfavor a lot / disfavor a bit / neutral / favor a bit / favor a lot - gammon-save attitude: disfavor a lot / disfavor a bit / neutral / favor a bit / favor a lot - double/nodouble attitude: very shy / shy / neutral / aggressive / very aggressive - take/pass attitude: very shy / shy / neutral / aggressive / very aggressive Another thing that could be interesting is to avoid applying the "presonality" all the time, on all the decisions: an "early doubler" will double early randomly in the match (e.g. with a given probability like 33% of the time, or something). MaX. _______________________________________________ Bug-gnubg mailing list Bug-gnubg@... http://lists.gnu.org/mailman/listinfo/bug-gnubg