Added a looser accuracy measurement for IPv6 OS training

[David Fifield <david () bamsoftware com>]

I was curious about how much of our cross-validation accuracy loss was
due to near misses: one Linux class being confused for another Linux
class, for example. In r35331 I added a second accuracy figure to the
output of train.py that considers it a match as long as two classes have
the same (vendor, family, generation, devicetype) tuple. About 10% of
the mismatches are such near misses.

$ ./train.py -c 100 -s nmap.set -g nmap.groups --scale > /dev/null
WARNING: No route found for IPv6 destination :: (no default route?)
Training.
Accuracy 69.3103448276
Same-nmapclass accuracy 79.6551724138
_______________________________________________
Sent through the dev mailing list
https://nmap.org/mailman/listinfo/dev
Archived at http://seclists.org/nmap-dev/