This release adds a test for HTML with scripting languages. This
follows a fixed format, and it is better to have a deterministic test
for it. The way the error is calculated when learning has changed.
Speed optimizations, an option for randomized learning in nn/learn,
and more compiler flags for better optimization have been added. A
fix has been made for a serious bug in the statistics gathering class
which caused it to not detect multiple occurrences of a reserved word.
