[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [RFR] templates://crm114/{crm114.templates}

Milan Zamazal wrote:
> CP> I'd recommend explaining (briefly) *what* makes the software
> CP> good....more than just say it is good..:-)
> Sure, this is why I've originally said that the description should
> retain the facts that crm114 is very accurate and learns fast; these
> facts are important for users looking for classifiers.  Anyway the
> original description should be changed, how about something like (I'm
> sure you can improve the wording):
> - Accuracy of the SBPH/BCR classifier has been seen in excess of 99 per cent,
> - for 1/4 megabyte of learning text. In other words, CRM114 learns, and it
> - learns fast.
> + Some of the categorization methods can classify with very high accuracy
> + after learning only relatively small amount of input text.

Meanwhile I'd been thinking along the lines of:

    CRM114 is not just another drop-in spam-filtering system; its
    Sparse Binary Polynomial Hashing methods give it the power to
    develop highly accurate Bayesian filters on very little training.

Or taking your version and just massaging it slightly:

    It has categorization methods capable of building a very high
    accuracy classifier from a relatively small supply of training

My third step ought to be to come up with a hybrid version, but
that's more than I can manage at the moment.  Any comments so far?
JBR	with qualifications in linguistics, experience as a Debian
	sysadmin, and probably no clue about this particular package

Reply to: