Re: Time Control for Blitz and 25 Minute Games

On Tue, 02 Dec 2003 18:04:58 +0100, Frank Hablizel
<[EMAIL PROTECTED]> wrote:

>Am Mon, 01 Dec 2003 22:17:06 +0000 schrieb David Richerby:
>
>> mondo  <[EMAIL PROTECTED]> wrote:
>>> Frank Hablizel wrote:
>>>> how should I calculate for a Blitz or a 25 minute game where I don't
>>>> know the exact number of moves?
>>>
>>> Do a google search on the topic of Iterative Deepening. This should help 
>>> answer most of your questions.
>> 
>> Iterative deepening allows you to fill some amount of time with useful
>> searching.  Frank wants to know how to calculate how much time to spend on
>> each move.
>> 
>> 
>> Dave.
>
>Yes Dave that's right. I know iterative deepening.
>
>I have i.e. the idea, to spend about 1/30th of the remaining time for each
>move. It's possible to multply that factor with another factor depending
>on the material that each player have.
>For the first approach this should work, but have anyone better ideas?
>

I like this basic idea.  It might also be good to include a factor for
the amount of advantage the program thinks it or its opponent has
based on the evaluation after the previous move. If the program is
clearly winning or clearly losing the game will probably not continue
for another 30 moves.  The extra time might come in handy to either
find the game-ending move or the saving resource.

Some other factors that could be included:
     * factor based on how close to best the opponents last response 
       was based on evaluation for last move.
     * factor based on the number of moves that are possible.  A     
       better measure might be the number of "reasonable" moves, but
       this could be suboptimal since sometimes an "unreasonable" move

       is would be discovered as best if only there was sufficient 
       time to analyze.

There are some exceptions to using this to determine time:
    1) when still in the program's opening book just use the time to 
        select the move from the book.
    2) when using an endgame tablebase just select the move
    3) when the program only has one move.

>But the best function is a function that use all the time and is flexible
>if the game is in a long range endgame.
>
>For this I think it's possible to use a table perhaps, because it should
>be possible to collect previous games for calculating the game
>length depending on similar material in other games?
>
>Have anyone use such a algorithm?
>
>Thanks in advance, 
>
>Frank
>