mars_doc.txt

统计模式识别算法包

                 Documentation for MARS driver code mars_dvr3.f 




The mars driver code mars_dvr3.f is modeled after the sample driver
routine supplied with the MARS code, but has some features which
make it easy to use.  This code must be compiled with the mars routines
contained in marslib.f.  This is done by typing:

g77 -g -o xmars mars_dvr3.f marslib.f -ff77 -fno-globals -Wno-globals

The executable will be the file xmars, which can be invoked by typing

xmars


In order to use the mars module with xtal, a copy of the executable must
be in the same directory containing xtal.   



The program xmars will ask for the following user parameters:


1.  Training set file name
The training data is expected to be in ASCII form with spaces or tabs as
column delimiters.  The last column is assumed to be the y variable.

2.  Number of samples in the training set
This informs the program how many rows are in the training set. The maximum
number of samples allowed is 10000.

3.  Number of predictor variables in training data.
This tells the program how many predictor or X variables are in the training
and test data.

4.  Test set file name
The test data is expected to be in ASCII form with spaces or tabs as
column delimiters.  The last column is assumed to be the y variable.

5.  Number of samples in the test set
This informs the program how many rows are in the test set.  The maximum number
of samples allowed is 10000.

6.  Maximum number of basis functions to use.
This parameter controls the complexity of the model and should be set between
1 and 100.

7.  Maximum number of degrees of freedom to use
This parameter should be between 0 and 9.  

8.  Maximum number of interactions to allow
The default value for this parameter is the number of predictor (X)
variables.  This parameter is used to limit the number of variables which
may enter in interaction terms of the basis functions.

9.  Type of spline fit
The default value for this parameter is 1 meaning piecewise-linear
splines are used.  If this is set to 2, then piecewise-cubic splines
are used.

10.  Name of output file used to hold run messages
The mars routines creates some useful and some not so useful
run time messages about the fit it produces.  If you would like
to save these, type in a file name here.  If you do not want
these saved, type the word 'none' here.

11. Name of output file used to hold fitted test set
After execution, this file will contain the test set data
only the y-values will be replaced with the estimates given by
MARS.  If you do not want this file, type 'none' here.

12. Name of the output file used to hold the RMS error
After execution, this file will contain the RMS error of
fitting the test set.  If you do not want this file, type 'none' here.

NOTE: the program does not produce any output to stdout, only
to the above output files.