computational power assertion

[Karl Wettin]

I've been thinking a bit about feature selection and have problems  
making up my mind when it comes to what sort of solutions makes most  
sense to initally focus on. If people have an unlimited number of  
nodes then complexity is no longer a big problem.

My guess is that people historically use nested subset ranking as the  
complexity is log N rather than N of exhaustive search wrappers. My  
guts tells me that the latter will produce a better result (given you  
also do Bonferroni correction or so) but who has an unlimited number  
of nodes?

The real question is hidden somewhere in between the lines and it  
doesn't only apply to feature selection.



         karl