Feature Classes

[Christine E. Suss]

Hi,

I am new to MaxEnt and I am having trouble defining feature classes in
an ecological sense. I understand the ecological interpretation of the
constraint for linear, quadratic and product that was provided in
Appendix 1: Details about features in Elith et al. 2011 Statistical
Explanation of Maxent but I do not understand the threshold and hinge
feature classes. Can someone provide me with a "non-mathematical"
description of the threshold and hinge feature?

Thank you!
Christine

[David Galbraith]

I'm not sure if this is correct, but I think of hinge features as being like threshold features, but different in that they do not imply a quantum leap (change from one state to another without experiencing the intermediate conditions along a gradient).  I think that with hinge features, part of an explanatory variable's range can have a linear relationship, while another portion of its spectrum could be defined by a nonlinear relationship of a much steeper or flatter linear relationship.

Not too sure.

Dave

--
You received this message because you are subscribed to the Google Groups "Maxent" group.
To post to this group, send email to max...@googlegroups.com.
To unsubscribe from this group, send email to maxent+un...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/maxent?hl=en.

[Rahlin, Anastasia]

Hi Christine, 

From my understanding, according to Elith et al 2011 "A statistical explanation of MaxEnt for ecologists" Appendix 1, 

Threshold is a "step function that allows a different response below the threshold to that above it," meaning that a covariate can have different constant (flat) responses depending on location on the map. Whereas hinge features, "the response above the knot or below the knot is linear with a positive or negative slope", meaning that a covariate can have different linear (not flat) responses depending on your location on the map.

Also I think if you use hinge features, using linear features becomes redundant, because linear features would fall out of hinge features anyway. As in, say you find only linear responses in your covariates, if you run the model only using hinge features, the hinge features would build those linear responses, because hinges are piecewise linear responses.

Take care,

Asya

--
Asya Rahlin

anastasi...@brown.edu
cell: 847.962.6775
Brown University

69 Brown Street #7020

Providence, RI 20912