OVERVIEW
========

This is CLAM, a prototype-based nearest neighbor learner (with
feature selection).

Rather than learn from all features and instances, CLAM prunes the
dull feature ranges then creates a small nuumber of prototypes (most
informative instances) using only those ranges.

In the following, anything in "quotes" is an option that can be set
on the command line.

INPUTS
======

CLAM accepts inputs in csv format. The first line must start with
a # symbol and is a list of column names. For numeric columns,
make the name start with upper case.  The last column is the class
column.

For example input files, see etc/data/*.csv.

Usage 
=====

Everything in CLAM is calls to the Makefile to generate new files from an
input .csv file.  The new files are one of:

.sym
----
The numeric columns of the .csv file replaced with a small bumber
of "BINS".  If the class column is numeric, then it is replaced
with 1 or 0 where 1 is union of all the "BEST" class bins and 0 is
the rest.  For examle, the file stem.sym is generated from stem.csv
via:   

	make stem.sym

.best
-----
This file lists the best ranges for each class.  The file stem.best
is generated from stem.csv via:   

	make stem.best

BUILT-IN EXAMPLES
=================

CLAM was built via TDD (test-driven development); i.e. by identifying
the next smallest, most useful, change to the code then implemetning
that delta, then writing a small test to demonstrate that delta.
At each step, an "egN" command was added to the system. Sometimes,
the output from "egN" was cached to etc/want/N.  As a result, this
distribution comes with numerous example commands already built in
and a regression suite to test everything is still working.

To run the one example, use

	make eg1

To run all the examples (which may be slow), use

	make egs 

To update the cached output