Subject: Marine Geospatial Ecology Tools (MGET) help
Text archives
From: | "Jason Roberts" <> |
---|---|
To: | <> |
Cc: | <> |
Subject: | RE: [mget-help] random forest in MGET |
Date: | Mon, 18 Mar 2013 11:15:31 -0400 |
Hi Margaret,
I fixed that problem last week in MGET 0.8a47, along with some other
problems relating to categorical predictor variables. It is no longer
necessary to write factor(categoricalpredictor) - categoricalpredictor. You
can just write factor(categoricalpredictor) and it should just work.
Please uninstall your existing MGET and try the latest one, 0.8a48, released
just now.
Also, MGET now supports both integer and character string categorical
variables. (Strings would be some text like Shelf, Slope, and Abyss.) If you
use strings, however, it makes prediction from rasters more complicated:
rasters cannot store strings (at least not trivially) so MGET requires you
to provide an integer raster instead, with integer codes representing the
strings. The codes must be numbered 1..n where n is the number of
categories. Codes are assigned in alphabetical order, e.g. Shelf, Slope, and
Abyss would have codes 2, 3, and 1, respectively. This is documented in the
Input Predictor Raster parameter of the tools named Predict <Model> From
Rasters.
This is not necessary for the tools called Predict <Model> From Table. Those
tools expect that if you used strings to fit the model, you will also use
strings to predict the model.
Best,
Jason
-----Original Message-----
From:
[mailto:]
Sent: Monday, March 18, 2013 10:15 AM
To:
Subject: RE: [mget-help] random forest in MGET
Hi, I am seeing the same error object'categoricalpredictor' not found if I
use
+ factor(categoricalpredictor) in the formula.
Just checking if there is any update on the horizon (ArcGIS10.1 version of
MGET) or if we should just use the workaround you suggested earlier for the
forseeable future.
i.e. + factor(categoricalpredictor) - categoricalpredictor
factor(categoricalresponse) seems to work OK.
thanks, Margaret
I fixed that problem last week in MGET 0.8a47, along with some other
problems relating to categorical predictor variables. It is no longer
necessary to write factor(categoricalpredictor) - categoricalpredictor. You
can just write factor(categoricalpredictor) and it should just work.
Please uninstall your existing MGET and try the latest one, 0.8a48, released
just now.
Also, MGET now supports both integer and character string categorical
variables. (Strings would be some text like Shelf, Slope, and Abyss.) If you
use strings, however, it makes prediction from rasters more complicated:
rasters cannot store strings (at least not trivially) so MGET requires you
to provide an integer raster instead, with integer codes representing the
strings. The codes must be numbered 1..n where n is the number of
categories. Codes are assigned in alphabetical order, e.g. Shelf, Slope, and
Abyss would have codes 2, 3, and 1, respectively. This is documented in the
Input Predictor Raster parameter of the tools named Predict <Model> From
Rasters.
This is not necessary for the tools called Predict <Model> From Table. Those
tools expect that if you used strings to fit the model, you will also use
strings to predict the model.
Best,
Jason
-----Original Message-----
From:
[mailto:]
Sent: Monday, March 18, 2013 10:15 AM
To:
Subject: RE: [mget-help] random forest in MGET
Hi, I am seeing the same error object'categoricalpredictor' not found if I
use
+ factor(categoricalpredictor) in the formula.
Just checking if there is any update on the horizon (ArcGIS10.1 version of
MGET) or if we should just use the workaround you suggested earlier for the
forseeable future.
i.e. + factor(categoricalpredictor) - categoricalpredictor
factor(categoricalresponse) seems to work OK.
thanks, Margaret
Archives powered by MHonArc.