Noel J. Bergman wrote:
if you are using the BayesianAnalysis mailet, keep in mind that the
in-memory corpus may slowly grow a little bit, checked mail after
checked mail. It is done on purpose to reduce the number of computations
of "degenerated tokens"
Should we look at using references to hold the in-memory corpus as a cache,
and recreate if necessary?
Good idea, but I would cache only a new corpus HashMap containing only
the "dynamic" tokens, as the main corpus itself is used all the time and
is heavy to load, while the dynamic tokens can be built very fast and
their "usefulness" is mostly for the current message. And because
of that, I would use a weak instead of a soft reference to such corpus.
I'll do it for 2.4 :-) .
Vincenzo
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
