RE: [computer-go] Tesuji solver

["Anders Kierulf" <anders@xxxxxxxxxxx>]

> Over the last years I've been working on a tesuji solver called MadLab
(based on the so-called lambda search algorithm). 
> [ ... ]
> Madlab comes with 997 tesuji problems extracted from real high-level
games.

Thanks for providing those capture problems, Thomas!

For the record, here is SmartGo's performance on those 997 problems:
- Solved correctly: 927 (with max number of nodes set to 300000).
- Failed: 70. The failures split up as follows:
    - Doesn't even try because it thinks the block has already escaped: 34
(e.g. problems 544, 924).
    - Searches but wrongly thinks block can escape: 9 (e.g. problems 492,
725).
    - Unknown, doesn't finish in 300k nodes (more nodes might help): 27
(e.g. problems 404, 944).

The total time for going through all problems was 11 min 42 sec (on a
Pentium 4 Xeon 2.8 GHz). Here's the breakdown of how quickly the problems
were solved:

Time        Solved     Percent of total
< 0.01s     270        27%
< 0.1s      567        57%
< 1s        801        80%
< 10s       927        93%
 
It would be interesting to see the results for other engines broken down the
same way.
 
Anders Kierulf
www.smartgo.com 


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/