Regularization Parameter

[Lucia Cenni]

Hello everyone. 

I'm Lucia Cenni and I am using the software MaxEnt for my Master thesis project. 

I a using presence-only occurrence records of an antelope in South Africa. 

I would like to understand something more about the regularization parameter and to know which one I should use (if needed). 

Can anybody help me? 

Thank you 

[Jamie M. Kass]

Please read the articles below. There is much information in these papers on regularization multipliers, feature classes, and most other Maxent things.

Phillips, S. J., & Dudík, M. (2008). Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31(2), 161-175.

Elith, J., Phillips, S. J., Hastie, T., Dudík, M., Chee, Y. E., & Yates, C. J. (2011). A statistical explanation of MaxEnt for ecologists. Diversity and Distributions, 17(1), 43-57.

Merow, C., Smith, M. J., & Silander, J. A. (2013). A practical guide to MaxEnt for modeling species’ distributions: what it does, and why inputs and settings matter. Ecography, 36(10), 1058-1069.

Jamie Kass
PhD Candidate
City College of NY

[Weverton Carlos]

Hi Lucia,

The regularization multiplier makes the constraints less rigorous. When you increase the regularization multiplier, the iterations and the gain decrease and the response curves become smoother.

I think the best way is to test several values and look at the response curves to choose the one that makes the most sense to the specie.

Best regards,

Weverton C. Trindade

State University of Ponta Grossa (Brazil)

[mjb...@york.ac.uk]

Just to follow up on the posts by Jamie and Weverton, I would also take a look at ENM tools and this reference: Warren, D. L., & Seifert, S. N. (2011). Ecological niche modeling in Maxent. The importance of model complexity and the performance of model selection criteria. Ecol Appl, 21(2), 335–342.

[Jamie M. Kass]

To follow up on Carlos' post, and just to clarify, increasing the regularization multiplier (RM) increases the penalty on allowable model complexity. Thus, higher values of RM will result in a simpler model, which means it will have fewer parameters with a non-zero beta coefficient. You can check the lambdas file to see which features of which variables have non-zero coefficients.

Increasing the number of feature classes does the inverse: it allows more model complexity (especially hinge, product, and threshold features). Thus, you should experiment with different combinations of these two parameters. This was the impetus for creating the R package ENMeval.

--

You received this message because you are subscribed to a topic in the Google Groups "Maxent" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/maxent/EFMseBJBPJw/unsubscribe.
To unsubscribe from this group and all its topics, send an email to maxent+unsubscribe@googlegroups.com.
To post to this group, send email to max...@googlegroups.com.
Visit this group at https://groups.google.com/group/maxent.
For more options, visit https://groups.google.com/d/optout.

--

-----------------------------------------------------------
Jamie M. Kass
PhD Candidate, Dept. of Biology
City College of New York, CUNY Graduate Center