Hi Michelle,

Thanks for your interest in MGET.

We agree that Random Forest models are very useful and would be a good addition to MGET. Our intent was to implement Random Forest tools similar to the CART/GLM/GAM tools this spring but some urgent projects came along and delayed us. Now it is unlikely we will have support for Random Forests before August or September. Do you have an active project that needs Random Forests? If so, we might be able to coordinate our work with you, for the opportunity to use your data as a test case.

On a related note, we will be adding tools for linear mixed effects models in July. I’m not sure if those would be useful for your situation, but if so, let me know and we can coordinate on that.

By default, MGET treats all numeric variables as continuous. To treat a variable as categorical, you should convert it to a factor in your model formula. For example:

Presence ~ SST + ChlDensity + Depth + factor(SubstrateType)

In this formula, SubstrateType will be treated as a categorical variable. Factors in R are basically equivalent to ordinal or nominal variables in other modeling environments.

Best regards,

Jason

Hello,

I recently downloaded your ArcGIS toolbox and am very impressed by how well the tools work and the great variety and quantity of useful tools that are made available. I have experimented a little with the GLM and CART tools provided in the toolbox and am excited about the potential for these tools. The question I have is this: is there or will there be the capability to perform Random Forest modeling with MGET?

The reason I ask is that random forest is also available in R and overcomes a lot of the problems associated with CART. There is a tool in R called ModelMap that does Random Forest modeling, but it would be much easier and more seamless to run RF models using MGET.

Also, something I have been unclear about is whether it is possible to identify predictor variables as categorical using the MGET modeling tools. Most of the programs I have used that employ statistical techniques that handle a variety of predictor variable types (i.e., continuous, ordinal, nominal) ask for categorical variables to be specifically identified. Are the modeling tools in MGET treating all predictor variables as continuous?

Michael

Michael Heilen, PhD, RPA

Research Director and Principal Investigator

Statistical Research, Inc.

571-248-2100 (voice)