news

卡内基梅隆大学MaCallum开发的文本分类系统

* New since libbow version 0.9

New classification methods have been added:  maxent, svm, active,
nbshrinkage. 

New libbow front-ends have been released:  crossbow (document
clustering), archer (AltaVista-style document retrieval).

Many bug fixes.


* New since libbow version 0.8.

The command-line arguments for setting test-train splits have
changed.  The -p option has been removed.  There are new options
--test-set and --train-set.

In the absence of any test-train split command-line options,
previously the default test-train split was 30% testing, 70% training.
Now the new default is 0% test, 100% train.  Thus, you must specify
some test-train split on the command-line for non-diagnostic
classification runs, such as runs with --test option.

The --split-seed option has been renamed to --random-seed.

Changed "infogain-vector" to "print-word-infogain".  
Changed "infogain-pair-vector" to "print-word-pair-infogain".  
Changed "weight-vector" to "print-word-weights".  
Changed "foilgain-vector" to "print-word-foilgain".


* New since libbow version 0.6.

The library and its front-ends now take GNU-style long-options.  For
example, to see the new rainbow options, execute "./rainbow --help".

NOTE: Several of the options have changed.  For example, `-R' should
be replaced with `-O'.  

WARNING: One of the options as reversed meaning!  Pre-version-0.7,
stemming was on by default; `-S' turned it off; now, stemming is
turned off by default; `-S' will turn it on.

WARNING: The data file format for models have changed.  You will have
to re-build any rainbow or arrow models you created with libbow
version 0.6.  Otherwise rainbow or arrow will crash when reading in
the old models.