Re: computer-go: Deep search = worst play ?

[Vincent Diepeveen <diep@xxxxxxxxx>]

What is your assumption in short?

Of course a deeper search will always help. If it doesn't
help then there is a bug in the programs search or eval design.

Usually evaluation functions are very bad debugged, so are searches,
apart from that there are typical qsearch problems and hash problems.

Most hash still with 32 bits, which is very bad. Even 64 bits is a 
bit little.

At 06:57 PM 9/6/00 +0200, you wrote:
>
>  With a wide enough search look-ahead usually improves evaluation.
>Of course evaluation should be doing what we intended to do: if you
>reverse the sign then look-ahead will also improve it ... finding
>even worst moves!
>  I think that the cases where deepening the search has bring out
>consistent worst play (of course sometimes it will change a move
>for anoter for bad) are due to a very narrow search, so too many
>reasonables moves are left out.
>  Of course, the narrower it is, the deeper it can be (time
>constraints) but if you left out the good answers no perfect evaluation 
>will help. 
>  Can the ones that have found that confirm/refute my assumption ?
>
>  Joan
>
>