CRM

Speedups were made, and an overflow problem for those with very large internally generated buffers has been fixed. The .css files for spam filtering have further learning in them and are now approaching a level of accuracy close to the author's own accuracy in determining spam v. nonspam

Improvements in spam tagging and mail system integration, better sparse spectra files for the Bayesian sorter, a cssutil utility to manage the css (Crm Sparse Spectra) files, and numerous bugfixes and de-idiotifications. Currently, the accuracy on the author's spam mix is 99.7%+, with feedback to the learning system required only when the system makes the occasional error.

This release contains the long-awaited sparse spectral mail filter, which is a learning filter based on Bayesian classification of phrases. It allows users to submit things they consider "blacklist", "whitelist", "spam", and "nonspam" just by mailing to themselves with a control password. Baseline setup files are included for user mode installation. Additional functionality and an embarrassingly large number of bugfixes are other bonuses.

More features were added. A bug in the WINDOW code
and another bug in the preprocessor were fixed.

New features include separate submatchings for
each captured variable, much faster windowing, an
explicit default window named :_data_window_:, and
latent code for auto-resizing window buffers as
needed. Examples are now more explicit and a whole
slew of regression tests are included.