Points for Pap318 comparing R-GARD to A-GARD

0. A main goal is to compare the two abovementioned GARD models and point
out similarities and differences. We note the constraints on such a
comparison, as the models are inherently different in some respects. But a
later part of the paper will also deal with analyses and properties of R-
GARD irrespective of such comparison.

1. To better compare R-GARD with A- GARD, IL will generate A-GARD carpets
using Euclidean distance instead of the dot product (H), which we currently
use.  We will also compare internally the difference between these two
metrics when applied to the same data.

2.  We will also calculate the compotype counts of A-GARD with kmeans
clustering by using Euclidean distance instead of dot product.  We should
check how this change affects the resulting number of compotypes. If for
some reason this does not work easily, we will try a more phenomenological
comparison.

3.  Choose carpets from R-Gard results for us to compare with similar
carpets from A-GARD.  These carpets should exhibit repeating compotypes,
drift, and composomes.

4.  Calculate and display statistics to emphasize the dependence of various
results on the parameters of R-GARD.  Example:  number of compotypes
dependency on split size.