0.9.9
-----
 - Added PerceptronRanker option. (Jason)


0.9.8
-----
 - Applied patch from Stephan Oepen that merges the functionality of
   score.cc into evaluate.cc. Removed score.cc from the
   repository. (Jason)

 - Updated for latest versions of PETSc (2.3.3) and TAO
   (1.9). Basically involved renaming a few variables and reordering
   the calls to some TAO functions. (Jason)

 - Added option --display_label_distributions (-x) to output the
   distribution the model assigns to test instances. For
   classification models, a list of tuples is produced, in which each
   tuple consists of the label and its probabality. For ranking
   models, a list of probabilities is produced, with each element
   corresponding to the appropriately ordered list of candidates in
   the test file. These can be easily read back into Python using the
   eval() function. (Jason)

 - Added option --max-it (-i) to specify the maximum number of
   iterations to perform when training a maxent model. (Jason)

0.9.7b
-----

 - Fixed bug in inducer.py due to Python indentation oversight
   introduced when adding support for ranking models in previous
   version. This will cause any models trained for simple
   classification to go awry (as can be seen by looking at the native
   TADM numeric format for event files generated by the previous
   version). (Jason)

0.9.7
----- 

 - Added Python support for ranking models. (Jason)
 - Added data sets for testing: tennis and parse_selection. (Jason)


0.9.6
----- 

 - Added perceptron estimation. (Jason)
 - Modularized the python interface code. (Jason)
 - Added Simple format for reading in events. (Jason)


0.9.5 
----- 

First TADM release, basically Rob Malouf's original code relicensed
under the Lesser GNU Public License.