Re: gtali computer play enhancements

["Andreas R." <andreasr gnome org>]

On Sat, 27 Jan 2007, Geoff Buchan wrote:
> I've been tinkering with some enhancements for computer play for gtali. So
> far I've implemented two things:
> 1. A command line argument, -c or --computer-test, which accepts an integer,
> and the program then runs that many games played only by the computer, and
> without the GUI. It outputs the score of each game to stdout, and then an
> overall average.
> 2. A different algorithm for determining what the computer will roll. Rather
> than relying on the "value" calculated in the present program, my new
> algorithm does a Monte Carlo simulation of each possible dice roll
> combination, and it then takes the combination that would result in the
> highest average score after 1000 rolls for each possible combination.
>
> This does appear to improve computer play a fair amount. I estimate the
> average score for the computer goes up about 20 points in regular mode, and
> 25 points in colors mode, with the new algorithm, but the new algorithm is
> more CPU intensive: running 100 "games" takes about a minute on my
> single-CPU 1.2 GHz laptop, whereas 100 "games" with the old algorithm takes
> about a second.
>
> It seems to me the best thing would likely be simply to improve computer
> play "quietly" - that is without changing the UI at all, but if we wanted
> to, we could either:
> 1. Simply retain the old computer play algorithm
> 2. Create a preference to allow users to decide between the older or newer
> play, and potentially also to control the number of trials in the Monte
> Carlo simulation (more trials is slower, but up to a point should improve
> play somewhat).
>
> What do you think would make the most sense ? I'd be happy to enter an
> enhancement request and patch for any of these possibilities. If you think
> someone else might be interested in this topic, please forward this message
> to him/her.

Hi! It would be interesting to try the new algorithm you propose for 
gtali. Please submit it to bugzilla. Then the patch can be applied to SVN 
during the 2.19.x development cycle.

 - Andreas