I posted something about a possible
indicator for volatility of a position (an attempt at
"explaining a cube decision"),
I put the discussion here below:
>> Massimiliano Maini wrote:
>> > Roughly (cubelessly) speaking, if the equity of the current
position is
>> > the weighted
>> > average of the equities of the subsequent postions in the
game tree,
>> > wouldn't the
>> > variance of such equities be an indicator of the volatility
?
Joerg picard replied this:
>How about displaying the number
of market losing rolls?
>
>If you want to get trickier you
could display cube statistics for 2-ply:
>x% of rolls leave you too good
>x% of rolls double drop
>x% double take
>x% no double
>x% double take for opponent
>x% double take for opponent
>x% too good for opponent
And Øystein replied this:
>> Yes, that's a good measurement
for the volatility. Of course it should
>> be 2-ply variance, since the that's what can happen until it's
your turn
>> again. I believe the doubling algorithm in Jellyfish worked with
at
>> volatility based on the 1-ply variance, and that algorithm wasn't
bad.
>>
>> Implementing such feature in GNU Backgammon should not be difficult,
>> however... hmmmm... where would we integrate this in the GUI?
and how do
>> we set the evalcontext etc. Small problems... can be solved....
>>
>> -Øystein
>
>I would put it it's in the Hint
dialog of a cube decision (and
>in the analysis pane, of course).
>
>2-ply variance is the logical thing
too look at, but even the 1-ply
>one indicates how volatile the position
is ...
>I would say that if cube eval is
2-ply, then the variance is 2-ply,
>else it is 1-ply.
>
>MaX.
It would be a bit of a new thing (at
least for gnubg), hence I would start with
something not too complicate and see
how it works. For example, take the
variance of the cubeful N-ply (0/1/2)
equity of the best move of each possible roll.