• A foolproof guide to Isight?

    From pepstein5@gmail.com@21:1/5 to All on Tue Apr 19 03:04:29 2022
    I'm pleased that I finally got round to reading Axel's paper.
    Despite my attempts to apply that method being wrong, I'm pretty
    confident I can apply it in the future. But I have the advantage
    of being incredibly good at mental arithmetic, so my case might
    be atypical.
    There are actually (at least) three Isight algos for money play.
    1) Simple Isight without using delta _ l
    2) The delta_l version
    3) Using both but having an algo to determine whether you use
    method 1 or method 2.

    More correctly, 3) is a class of algos rather than a single algo.

    Suppose we have a non-contact position and a player knows her mental
    arithmetic ability. 100% means every computation is instant.
    50% ability means that pip counts of approx 100 for each side takes
    two minutes to evaluate.

    Given the player's counting ability, it would be useful to have a clear algo. Presumably a (theoretical) 100% player should always use the delta_l version?

    For non-100% players, it would be useful (in my opinion) to flesh out
    the algo for a range of mental arithmetic abilities.
    For example, suppose the full 80 - l/3 + 2 * delta_l computation
    takes 5 minutes?

    It could also be useful to have simplifications such as: If you lead
    by X pips and Y is true, it's D/P anyway.

    Paul

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Axel Reichert@21:1/5 to peps...@gmail.com on Tue Apr 19 13:10:13 2022
    "peps...@gmail.com" <pepstein5@gmail.com> writes:

    There are actually (at least) three Isight algos for money play.
    1) Simple Isight without using delta _ l
    2) The delta_l version
    3) Using both but having an algo to determine whether you use
    method 1 or method 2.

    1 and 2 give EXACTLY the same cube action in a money session. I prefer
    the more general Delta l version, because it will give you a winning percentage, which is required for match play. Then you "just" need to do
    some match equity calculation to come up, say, with your cash point
    (typically you want to double close to it) and act accordingly.

    It could also be useful to have simplifications such as: If you lead
    by X pips and Y is true, it's D/P anyway.

    I think a pretty robust recommendation by Robertie is that it is a
    double if you have a smaller pipcount AND a smaller number of checkers
    on board.

    Best regards

    Axel

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From pepstein5@gmail.com@21:1/5 to peps...@gmail.com on Tue Apr 19 04:41:19 2022
    On Tuesday, April 19, 2022 at 12:33:19 PM UTC+1, peps...@gmail.com wrote:
    On Tuesday, April 19, 2022 at 12:10:15 PM UTC+1, Axel Reichert wrote:
    "peps...@gmail.com" <peps...@gmail.com> writes:

    There are actually (at least) three Isight algos for money play.
    1) Simple Isight without using delta _ l
    2) The delta_l version
    3) Using both but having an algo to determine whether you use
    method 1 or method 2.
    1 and 2 give EXACTLY the same cube action in a money session. I prefer
    the more general Delta l version, because it will give you a winning percentage, which is required for match play. Then you "just" need to do some match equity calculation to come up, say, with your cash point (typically you want to double close to it) and act accordingly.
    It could also be useful to have simplifications such as: If you lead
    by X pips and Y is true, it's D/P anyway.
    I think a pretty robust recommendation by Robertie is that it is a
    double if you have a smaller pipcount AND a smaller number of checkers
    on board.

    Best regards

    Axel
    Thanks. I can't resist testing it, now that I have more info.
    I'm going to try and be a bit sneaky and trip Isight up a little bit.
    Let's make the position a bit atypical by giving the opponent just
    a single checker on her 6 point. This maximises her recube vig.
    I'll be on roll with two checkers on my two point.
    This is well-known as redouble/take.
    But, since the opp's recube vig is so high, she can take even though
    her winning probability is small -- will this fool Isight?
    I have a gap on my six point that she doesn't have.
    So my count is 5. Hers is 6.
    5 + 5/6 - 6 is negative so D/P which is wrong.

    80 - 5/3 + 2 * 1 is bigger than 76 which is a wrong D/P as before.
    Yes, the D/P is wrong. But maybe the probability is fine.
    The probability of me winning if the game is played to the end
    is 1 - (5/18 * 3/4) = 57/72 = 19/24 = 79%. 80 - 5/3 + 2 = 80 1/3 which ain't a bad approx.

    I don't see the reason why the two algos would always give the same cube actions but I'll take your word for it.
    Oh, ok. I think I get it. You start with an 80 year old man (call him Bob) and then assess how well he gets on
    with his friends aged 68, 70 and 76. So there are differences involved and it probably translates to the 6, 5, 2 pattern when
    you do the manipulations.

    Paul

    I knew I'd get these wrong at first. I forgot my extra checker. This makes the method better but I'm not sure whether it will work.
    Both our counts are in fact 6. 6 + 6/6 - 6 = 1. Still D/P.

    How about the probability? 80 - 6/3 + 1 = 79%. Wow, that is pretty much perfection!

    Paul

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From pepstein5@gmail.com@21:1/5 to Axel Reichert on Tue Apr 19 04:33:18 2022
    On Tuesday, April 19, 2022 at 12:10:15 PM UTC+1, Axel Reichert wrote:
    "peps...@gmail.com" <peps...@gmail.com> writes:

    There are actually (at least) three Isight algos for money play.
    1) Simple Isight without using delta _ l
    2) The delta_l version
    3) Using both but having an algo to determine whether you use
    method 1 or method 2.
    1 and 2 give EXACTLY the same cube action in a money session. I prefer
    the more general Delta l version, because it will give you a winning percentage, which is required for match play. Then you "just" need to do
    some match equity calculation to come up, say, with your cash point (typically you want to double close to it) and act accordingly.
    It could also be useful to have simplifications such as: If you lead
    by X pips and Y is true, it's D/P anyway.
    I think a pretty robust recommendation by Robertie is that it is a
    double if you have a smaller pipcount AND a smaller number of checkers
    on board.

    Best regards

    Axel

    Thanks. I can't resist testing it, now that I have more info.
    I'm going to try and be a bit sneaky and trip Isight up a little bit.
    Let's make the position a bit atypical by giving the opponent just
    a single checker on her 6 point. This maximises her recube vig.
    I'll be on roll with two checkers on my two point.
    This is well-known as redouble/take.
    But, since the opp's recube vig is so high, she can take even though
    her winning probability is small -- will this fool Isight?
    I have a gap on my six point that she doesn't have.
    So my count is 5. Hers is 6.
    5 + 5/6 - 6 is negative so D/P which is wrong.

    80 - 5/3 + 2 * 1 is bigger than 76 which is a wrong D/P as before.
    Yes, the D/P is wrong. But maybe the probability is fine.
    The probability of me winning if the game is played to the end
    is 1 - (5/18 * 3/4) = 57/72 = 19/24 = 79%. 80 - 5/3 + 2 = 80 1/3 which ain't a bad approx.

    I don't see the reason why the two algos would always give the same cube actions but I'll take your word for it.
    Oh, ok. I think I get it. You start with an 80 year old man (call him Bob) and then assess how well he gets on
    with his friends aged 68, 70 and 76. So there are differences involved and it probably translates to the 6, 5, 2 pattern when
    you do the manipulations.

    Paul

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From pepstein5@gmail.com@21:1/5 to peps...@gmail.com on Tue Apr 19 04:43:16 2022
    On Tuesday, April 19, 2022 at 12:41:21 PM UTC+1, peps...@gmail.com wrote:
    On Tuesday, April 19, 2022 at 12:33:19 PM UTC+1, peps...@gmail.com wrote:
    On Tuesday, April 19, 2022 at 12:10:15 PM UTC+1, Axel Reichert wrote:
    "peps...@gmail.com" <peps...@gmail.com> writes:

    There are actually (at least) three Isight algos for money play.
    1) Simple Isight without using delta _ l
    2) The delta_l version
    3) Using both but having an algo to determine whether you use
    method 1 or method 2.
    1 and 2 give EXACTLY the same cube action in a money session. I prefer the more general Delta l version, because it will give you a winning percentage, which is required for match play. Then you "just" need to do some match equity calculation to come up, say, with your cash point (typically you want to double close to it) and act accordingly.
    It could also be useful to have simplifications such as: If you lead
    by X pips and Y is true, it's D/P anyway.
    I think a pretty robust recommendation by Robertie is that it is a
    double if you have a smaller pipcount AND a smaller number of checkers
    on board.

    Best regards

    Axel
    Thanks. I can't resist testing it, now that I have more info.
    I'm going to try and be a bit sneaky and trip Isight up a little bit.
    Let's make the position a bit atypical by giving the opponent just
    a single checker on her 6 point. This maximises her recube vig.
    I'll be on roll with two checkers on my two point.
    This is well-known as redouble/take.
    But, since the opp's recube vig is so high, she can take even though
    her winning probability is small -- will this fool Isight?
    I have a gap on my six point that she doesn't have.
    So my count is 5. Hers is 6.
    5 + 5/6 - 6 is negative so D/P which is wrong.

    80 - 5/3 + 2 * 1 is bigger than 76 which is a wrong D/P as before.
    Yes, the D/P is wrong. But maybe the probability is fine.
    The probability of me winning if the game is played to the end
    is 1 - (5/18 * 3/4) = 57/72 = 19/24 = 79%. 80 - 5/3 + 2 = 80 1/3 which ain't a bad approx.

    I don't see the reason why the two algos would always give the same cube actions but I'll take your word for it.
    Oh, ok. I think I get it. You start with an 80 year old man (call him Bob) and then assess how well he gets on
    with his friends aged 68, 70 and 76. So there are differences involved and it probably translates to the 6, 5, 2 pattern when
    you do the manipulations.

    Paul
    I knew I'd get these wrong at first. I forgot my extra checker. This makes the method better but I'm not sure whether it will work.
    Both our counts are in fact 6. 6 + 6/6 - 6 = 1. Still D/P.

    How about the probability? 80 - 6/3 + 1 = 79%. Wow, that is pretty much perfection!

    Paul

    Sorry, the probability is 80 - 6/3 + 2 * 0 = 78 instead of 79.

    Paul

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Axel Reichert@21:1/5 to peps...@gmail.com on Tue Apr 19 16:49:22 2022
    "peps...@gmail.com" <pepstein5@gmail.com> writes:

    I don't see the reason why the two algos would always give the same
    cube actions but I'll take your word for it.

    This is by design. Explanation starting on page 24. Essentially Bower's interpolation generalized.

    Oh, ok. I think I get it. You start with an 80 year old man (call
    him Bob) and then assess how well he gets on with his friends aged 68,
    70 and 76. So there are differences involved and it probably
    translates to the 6, 5, 2 pattern when you do the manipulations.

    Roughly. (-:

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)