investigate grid search quality for random forest

RockStarCoders / alienMarkovNetworks

Using MRFs and CRFs for computer vision problems.

21 stars 9 forks source link

investigate grid search quality for random forest #24

Closed jsherrah closed 10 years ago

jsherrah commented 11 years ago

did we put enough choices? are there other parameters we should add?

are the cross-val folds etc right?

jsherrah commented 11 years ago

All the params are there. The folds is always a trade-offf: the more the better, but takes longer.

jsherrah commented 10 years ago

Overall the grid search looks ok. I've changed it to fix the number of trees since bigger is always better.

The main thing to look for was whether it's always picking parameters at the end of the range, which means we're not searching a large enough range of values. I expanded the range on a few to extreme values, and these were selected as best, but we can't really push it out any further. Consider it done.