This week, I worked on improving the interface to the classifier code and
incorporating a simple grid averager.  Right now, the code takes in
evidence grids sequentially and can classify/manipulate the grids with
three separate algorithm:

1)  BEST -  The incoming grid is matched with all previous instances.  The
	best matching instance is then used to determine the class of the
	incoming grid.  However, if even the best match is below a pre-set
	threshold, a new class is created instead.

2)  AVERAGE - The incoming grid is matched with all previous instances.  The
	best match among the preexisting CLASS (category/places/etc.), i.e.
	all the average match values of each class, is used to determine the
	category for a class.  If this averaged match value is below a pre-
	set threshold, a new class is created instead.

3)  BLEND - The incoming grid is matched with all previous instances.  If the
	threshold is high enough, the new instance and the best match is
	then averaged together, and this averaged grid replaces the original
	best matching instance.  If the threshold is not high enough, the
	new instance is then merely added to the database, unchanged.

Currently, I am working out a small problem with the third algorithm
(apparently, Karl didn't anticipate the grids might be changing, so I need
to track down a memory bug); the first two algorithm is working quiet well.
The next step is to conduct actual experiments that compare the three
techniques.

Questions that remains:
1) Is there some systematic way of finding a threshold?

2) Should there be two threshold used for the third algorithm?  One for use
 in determining when a good match is made, and another is used to determine
 if the match is REALLY good that we should do matching.

I anticipate I should be able to track down the bug by tonight (hopefully).
Experiments with Karl's preexisting grids will only take a few days.  I am
still not sure the status of new grids/experiments.  Even if new grids
are generated and use, I don't anticipate that it will take more than until
the end of next week (September 8th), which still leaves a little less than
two weeks for any future work.  Any ideas?