[seek-dev] Re: garp with kepler
Pereira, Ricardo Scachetti
rpereira at ku.edu
Fri Jan 16 20:12:47 PST 2004
Hi, Chad and Rich.
Yes, that is the spirit. Usually we get a really well known species and run through GARP and analyse the results.
Why don't you guys start with this sample Town prepared for me some time ago? The dataset and input data points are at http://www.cria.org.br/~ricardo/SEEK/SampleDataset.zip and the resulting maps generated by GARP are at http://www.cria.org.br/~ricardo/SEEK/resulting_maps.zip.
Since you are using GARP library directly, you will need to reformat the data points file. Probably all you need to do is to get only the coordinates off the file and feed the coordinates to GARP as a string with all coordinate pairs concatenated. Make sure LONGS come before LATS in each pair or your points will be transposed (Long, Lat corresponds to X, Y).
That is a really known bird species that should occur everywhere west of the Rocky mountains.
I don't know if that will help, but you can use the complete DesktopGarp modeling package (that essentially is a VB application on top of the GARP libraries you have). You can download it at http://www.lifemapper.org/desktopgarp/. I generated those results with that sample dataset that ships with the package.
Sometimes I also get the "all blank" problem you are having. This probably mean that there is something wrong in the way the input data to GARP. It usually happens when you switch pairs of coordinates (lats, longs instead of long, lat).
Try that at your end. I'll try to set up Kepler to see what goes on with it at my end.
Have you all a good holiday!
Regards,
Ricardo
-----Original Message-----
From: Rich Williams [mailto:rwilliams at nceas.ucsb.edu]
Sent: Fri 1/16/2004 7:32 PM
To: Chad Berkley; Pereira, Ricardo Scachetti
Cc: Deana Pennington; Matt Jones; Rich Williams; seek-dev at ecoinformatics.org
Subject: RE: [seek-dev] Re: garp with kepler
Hi Ricardo -
The output with the test data that Chad is currently using is all light
green (all present) on the continents and all black on the oceans. Since we
don't know what the current test data should produce, we don't know whether
the 'all present' output is correct or a bug. It would be useful to have a
set of input data that is known to produce something interesting (something
other than all present!) and make sure that it produces a qualitatively
similar map when run in kepler.
Rich
> -----Original Message-----
> From: seek-dev-admin at ecoinformatics.org
> [mailto:seek-dev-admin at ecoinformatics.org]On Behalf Of Chad Berkley
> Sent: Friday, January 16, 2004 3:00 PM
> To: Pereira, Ricardo Scachetti
> Cc: Deana Pennington; Matt Jones; Rich Williams;
> seek-dev at ecoinformatics.org
> Subject: [seek-dev] Re: garp with kepler
>
>
> Hi Ricardo,
>
> Sorry, I should've known that the word "exact" doesn't really work with
> genetic algorithms. I guess I just needed what you just provided...that
> is, an overview of what the output should (kind of) look like. If you
> could get it running and check out the output, that would be great. let
> me know if you need any help getting it going.
>
> chad
>
> Pereira, Ricardo Scachetti wrote:
> > Hi, Chad.
> >
> > It is great that you guys managed to make GARP run in Kepler.
> > I'll follow your instructions and will take a look at what you
> > have. To know what is going on inside GARP and help you with testing I
> > need to know what methods you are using to feed data to GARP.
> > I don't know what you mean by the word EXACT in your sentence:
> >
> > "...I need to know exactly what the output should be for a given
> > input...".
> >
> > If you want to give GARP an input and compare each bit of what
> > you get as result with a standard output, I would suggest you to forget
> > it.
> > The most difficult aspect of GARP is to test the results. To
> > make a long story short, those are the main reasons for that:
> > 1) GARP is a genetic algorithm and it takes a random walk at
> > some point to get rid of locally optimal solution. That means if you run
> > it twice with the same input data, you will get equivalent results that
> > are different;
> > 2) Because of the way the models are expressed (as rulesets),
> > GARP to come up with different rulesets that basically describe the very
> > same (or approximate) ecological niche. You can have a rule set with 5
> > rules that is essentially the same as another one with 10 rules.
> > That doesn't mean that the tested. For know (before I can get
> > back to you) you should get a map on the output that has the oceans in
> > black and the continents colored with dark green (species predicted
> > absent) light green (species predicted present) and gray (neither one,
> > no prediction there).
> > So I'll install it in my workstation and will take a look inside
> > and will get back to you to help you out (if you think this approach is
> > feasible).
> > What I've done is to look more carefully inside the algorithm to
> > see if things are going fine (mostly manually!?!). Having experts look
> > at the results is also a possibility. I wish there was a better way to
> > test GARP, and if you find a better one, please let me know.
> > Also, you might ask David Stockwell about this issue as well.
> > I'll get back to you.
> > Regards,
> >
> > Ricardo
> >
> >
> > PS. I'm on Central Time (Kansas) and about to call it for today, but
> > I'll try to get back to you ASAP.
> >
> >
> > -----Original Message-----
> > From: Chad Berkley [mailto:berkley at nceas.ucsb.edu]
> > Sent: Friday, January 16, 2004 4:03 PM
> > To: Ricardo Scachetti Pereira
> > Cc: Deana Pennington; Matt Jones; Rich Williams
> > Subject: garp with kepler
> >
> > Hi Ricardo,
> >
> > I've gotten GARP to run inside Kepler, but I need to know exactly what
> > the output should be for a given input. I've been using World.dxl as
> > the EnvLayerSet, and the test data file "DataPoints.txt" as the
> > DataPoints file (you sent me a zip file a while back with a bunch of
> > test data...both of these files came from that zip file). When I run
> > garp I get a sparse matrix of mostly -9999 values and a bitmap file that
> > appears just to be a picture of the contintents. Could you give me a
> > good scenario to run with a description of the output so I can do some
> > more testing?
> >
> > If you would like to try this out yourself in kepler, here are the
> > steps:
> >
> > windows and linux
> > -----------------
> > 1) download ptolemy II 3.0.2 (source version) from
> > ptolemy.eecs.berkeley.edu and unzip it.
> > 2) checkout kepler from cvs.ecoinformatics.org
> > 3) copy the build-ptolemy.xml file from the kepler directory to the
> > ptolemy directory
> > 4) set the environment variables PTII=<ptolemydir>, KEPLER=<keplerdir>
> > and JAVA_HOME=<javadir>
> > 5) go into the ptolemy directory and run 'ant install' (ptolemy should
> > compile and build a jar file)
> >
> > windows
> > -------
> > 6) copy the garp.dll and libexpat.dll files from kepler/lib to
> > windows/system32 (if windows is not installed in c:\windows, you must
> > chage the java library path in the win.prep target in the
> > kepler/build.xml file).
> >
> > linux
> > -----
> > 6) go into the kepler/src/cpp/garp directory and follow the steps in the
> > README file
> >
> > windows and linux
> > -----------------
> > 7) go into the kepler directory and run 'ant run' then choose the GARP
> > pipeline link when the kepler main screen comes up.
> >
> > * note you need ant verion 1.5.4 and java 1.4.1 or greater. for linux
> > you need gcc version 3.2 or greater.
> >
> > let me know if you have any questions.
> >
> > thanks,
> > chad
> >
> >
> > --
> > -----------------------
> > Chad Berkley
> > National Center for
> > Ecological Analysis
> > and Synthesis (NCEAS)
> > berkley at nceas.ucsb.edu
> > -----------------------
> >
>
>
> --
> -----------------------
> Chad Berkley
> National Center for
> Ecological Analysis
> and Synthesis (NCEAS)
> berkley at nceas.ucsb.edu
> -----------------------
>
> _______________________________________________
> seek-dev mailing list
> seek-dev at ecoinformatics.org
> http://www.ecoinformatics.org/mailman/listinfo/seek-dev
More information about the Seek-dev
mailing list