Test sample file

[Mariana Martinez-Morales]

Hi all Maxent users:

I am new to Maxent and trying to model the distribution of a rare and
endangered plant species, but have some very basic methodological
questions, hopefully somebody can help me.
I have a dataset of 68 points of presence and want to divide it in two
for i) training and ii) testing the model. I used the training dataset
to run the model and
select the testing file in the ¨Test sample file¨option in settings.
However, when I tried to run it, a window with a ¨Skipping testfile
because it has 0 test samples¨ appears and it stops running.
So, my question is
1) how should I select the ¨test file¨?
2) should the info from the test file also be included in the sample
file for running the model (I already tried unsuccessfully)
3) or in some specific format?
Pretty sure is something very simple, I already google it but didn´t
find nothing helpful!!
I´ll be very grateful for your help

Thank you very much
Mariana

[semra yalcin]

Hi Mariana,

There are two options for using test data in the model. 1) to use "Random test percentage" on the tab of settings. In this option, only sample file is used. Maxent selects test points randomly from the sample file according to percentage indicated. 2) to use "Test sample file" as you mentioned. Two CSV file is needed; sample and test. Headings (species, lang, lat) and species name should be same in both file. In this option, you divide points manually.

If your presence data is gathered in same way-same survey, you can use first option and adjust 50% of sample point to set as test points.


Regards,

Semra

--
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.

[Mariana Martinez-Morales]

Hi Semra:

Thank you very much for your quick and very good response!! I have run
it and I obtain a AUC for testing data of 0.993, which is pretty high.
Can I ask you whether this test is enough to report or do I need other
tests?
Thanks
Mariana

[Maxin]

Hello Mariana

Based on results from our work and from perspectives presented in
several publications, we recommend caution in placing too much faith
in the AUC figure as an assessment of the accuracy of the model.

Suggest checking out the following paper for a start:
Lobo JM, Jimenez-Valverde A and Real R. (2008) AUC: a misleading
measure of the performance of predictive distribution models. Global
Ecology and Biogeography 17: 145-151.

Regards
Rodney

On Mar 2, 4:06 pm, Mariana Martinez-Morales

> >http://groups.google.com/group/maxent?hl=en.- Hide quoted text -
>
> - Show quoted text -

[Mariana Martinez-Morales]

Thanks Rodney:
Yeahhh I´ve read this paper, and I´m using other measurements of
accuracy (such as Omission error, Commission index used in several
papers) to compare models. However, I´d like to use AUC as another
measurement to choose the best model. Do you have any idea on how to
obtain AUC for a model (which I run in GARP). I have the confusion
matrix, but do not know how to obtain the AUC from scratch?
Can anyone else help me?

Mariana

[AndrewG]

I beleive you calculate the ROC (and subsequently the AUC) by
calculating the ommission/commision errors for a range of thresholds
between 0 and 1. Although I dont have a script for doing this so that
is probably not much help... :|
Andrew
P.S. I also remember reading somewhere, that a species with a very
restricted distribution will often return a very high AUC.

[John Baumgartner]

Check out the r packages 'dismo' (see examples in ?evaluate) or 'ROCR'.

dismo: http://cran.r-project.org/web/packages/dismo/index.html
ROCR: http://cran.r-project.org/web/packages/ROCR/index.html

[Mariana Martinez-Morales]

Thanks Andi and John for your messages: Your opinion was very helpful,
now I understand better, I´ll check the the r packages recommended and
let you know what happen.

Cheers
Mariana