[computer-go] Caching of local search in NeuroGo?

[Markus Enzenberger]

On Tuesday 08 August 2006 21:58, Chrilly wrote:
> Suzie searches at the moment also only ladders. But somewhat generalized,
> there can be a branching. The criterion is that the prey has 2 liberties if

yes, the branching factor can be greater than 1, but these cases are 
relatively rare, the average branching factor is very close to 1. My 
incremental updating of blocks is also optimized for the creation of 
single-stone blocks and appending of a stone to an existing block, as long as 
no blocks are merged, I can do 100KNodes/sec.

> I assume in case of a Neural Net the relations are different and the local
> search consumes only a small fraction of the time.

that's right. Ladder search was the bottleneck in my first version of NeuroGo 
(the one with the dynamic graph architecture), but in the current version the 
network part is much slower, it evaluates only 200-300 global nodes a second, 
so I would gain nothing by speeding up the ladder searches.

- Markus