CHANGELOG.txt

Version 0.5

Added candidate evaluation to "gismu_score.py".

When generate candidates with custom shapes, excluding forbidden
consonant triples. Also excluding internal sequences which could be
mistaken for "ccvcv" gismu.

Deduplication and serialization of candidates are now options which are
disabled by default.

----

Version 0.4

Replaced mlpy implementation of LCS (longest common subsequence)
algorithm with a pure-python implementation to simplify
installation and enable compatibility with python implementations
such as pypy and jython.

By default, only generating candidates from letters which appear in input
words. This substantially reduces processing time by avoiding the calculation
of scores for less-relevant candidates.

Added command-line options to "gismu_score.py" to customize language weights,
gismu candidate shape, gismu alphabet, and number of scoring worker threads.

Output of "gismu_score.py" / input of "gismu_best.py" now uses python
"marshal" format due to OutOfMemory errors with pickle/cPickle under
jython.

The gismu candidates DBM database used by previous versions of the script
("candidates.db") is no longer created or consulted.

----

Version 0.3

Corrected algorithm to factor in the length of input words per CLL: "The scores
were divided by the length of the source-language word in its Lojbanized form"

----

Version 0.2

Integrated mlpy implementation of LCS (longest common subsequence) algorithm
for better performance.

----

Version 0.1

Arnt Richard Johansen's original implementation as downloaded from:

  http://www.lojban.org/tiki/gismu+scoring+algorithm