ENMTools comparing models

[DavidBecas]

Hi all,

I've done a few models using Maxent and would like to compare them
using ENMTools. However, because I've used cross-validation with 5
replicates I don't have the lambdas file for the average but 5 lambdas
files, one for each replicate. My idea was to compare several averaged
models all runned with the same cross-validation with 5 replicates but
because there's no average lamdbas file it seems this is not possible.
Does anyone have an idea on how to compare this average models? Is it
cross-validation with replicates not a good option for modeling?

Thanks,

David

[Jim777]

Hi David,

I have exactly the same problem.  Did you have any luck working this out?

Regards,

Jim

[DavidBecas]

Not yet!

Let me know if you get something!

Cheers,

David

[Jim777]

Hi David,

I couldn't work out how to average the lambda files so to do the ENM tools model selection analysis the choices are to do only 1 cross validation replicate or, if more than one is done, to include the asc and lambda files from all replicates in the model selection analysis (remember to tick the 'write output grids' otherwise only the average asc file will be outputted).  I am just about to try the second option to to see how it goes.  Hopefully a way to average lambda files come along soon.

[Hackett]

I tried averaging the five sub-models' AIC and chose the "best", wandering wether that's a right way...

Hackett

[John Baumgartner]

Models will likely have included different features, and I think averaging coefficients across models really doesn't make sense. You could use cross-val to assess sensitivity to choice of test/training set and then fit to the full data set as well. 

On 02/11/2012, at 1:01 PM, Hackett <lyt...@gmail.com> wrote:

I tried averaging the five sub-models' AIC and chose the "best", wandering wether that's a right way...

Hackett

--
You received this message because you are subscribed to the Google Groups "Maxent" group.
To view this discussion on the web visit https://groups.google.com/d/msg/maxent/-/QtElv-bzcCkJ.
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.

[Mar]

Hi David, 

I wonder if you have eventually found a solution to this problem of replicate seelction.

If yes, could you please let me know.

Best

mar

[Jocelyn Akins]

Hi Mar,

Did you get any resolution on the issue. Would be great to know.

Thanks,
Jocelyn

[Jordan DiNardo]

Hi David, 

I see its been a while since you first posed this question. Have you found a solution and if so could you share?

Thanks,

Jordan

On Monday, March 12, 2012 at 10:40:47 AM UTC-7, DavidBecas wrote:

[Jamie M. Kass]

Reading all the posts in this thread did not provide enough information for me, and likely for others too. Can you please detail the particular problem you're having? When one says they are "comparing models with ENMTools", what kind of comparison is this, and what exactly does one want to average together from the lambdas files? Thanks.

Jamie