BRT tuning in SAHM

[Aaron Sidder]

Hello,

I was hoping someone could clarify something from the BRT documentation.  It states, " It incorporates advanced algorithms for tuning the model settings..."  Does this mean that the algorithm runs through different values for parameters such as BagFraction, LearningRate, and MaximumTrees and selects the best model based on those tuning alterations?  I'm trying to decide if I need to do it manually or if that was incorporated in the module, thanks.

Aaron 

[Jarnevich, Catherine]

HI Aaron,

So tuning does occur in the model itself automatically.  However, I always check to see if the model is overfit and often find that it is.  To do this I compare the cross-validation statistics to the training statistics.  If the values are very different, I go back to the learning rate and tree complexity, varying these using the parameter explore option in VisTrails to optimize them to come up with the highest assessment metrics that are also close together for the training and CV.

I hope that helps.  There's a good explanation in the Elith paper on BRT that I believe is referenced in the documentation.
Catherine

--
You received this message because you are subscribed to the Google Groups "VisTrails SAHM" group.
To unsubscribe from this group and stop receiving emails from it, send an email to vistrails-sah...@googlegroups.com.
To post to this group, send email to vistrai...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Catherine Jarnevich
Research Ecologist
U.S. Geological Survey
Fort Collins Science Center
2150 Center Ave Bldg C
Fort Collins, CO 80526
970-226-9439