train.Rd
This function runs all of the steps in the analysis that depend only on the
input data. The results are saved in a structure of class fldgen
.
This structure contains everything the emulator has learned about the model.
Therefore, it can be saved and reloaded in future sessions to bypass the
analysis step.
train( dat, latvar = "lat", lonvar = "lon", meanfield = pscl_analyze, record_absolute = FALSE )
dat | A single directory name, or a list of netCDF files. If a directory name is given, all netCDF files in the directory will be used. |
---|---|
latvar | Name of the latitude coordinate variable in the netCDF files.
Normally this is |
lonvar | Name of the longitude coordinate variable in the netCDF files. |
meanfield | Function to compute the mean temperature response field. The default is a linear pattern scaling algorithm. |
record_absolute | If |
A fldgen
object.
The steps run by this procedure are read.temperatures
,
pscl_analyze
, and eof_analyze
. Additionally, the
global mean temperature and the FFT of the EOF decomposition are calculated
and stored. For backward compatibility, all of those functions remain part
of the public interface.
At present the field generation functions still take as arguments the
individual components of the structure; however, in the next version they
will expect a fldgen
object, which can be constructed manually using
fldgen_object
.
The names of the input files are recorded in the infiles
field of the
returned object. This is for the your information only; none of the code
here does anything with the recorded filenames. You have the choice of
recording either absolute filenames (useful for determining exactly which
dataset was used, if you have a lot of similarly named files) or relative
filenames (useful if you might distribute a
saved model to other users who might not have the same directory structure as
you).