Version 0.9.11

• Fixed all known remaining compatibility issues with Python 3.
• Fixed bug in skll.metrics.kappa which would raise an exception if full
  range of ratings was not seen in both y_true and y_pred. Also added a
  unit test to prevent future regressions.
• Added missing configuration file that would cause a unit test to fail.
• Slightly refactored skll.Learner._create_estimator to make it a lot
  simpler to add new learners/estimators in the future.
• Fixed a bug in handling of sparse matrices that would cause a crash if
  the number of features in the training and the test set were not the same.
  Also added a corresponding unit test to prevent future regressions.
• We now require the backported configparser module for Python 2.7 to make
  maintaining compatibility with both 2.x and 3.x a lot easier.

Version 0.9.10

• Fixed bug introduced in v0.9.9 that broke "predict" mode.

Version 0.9.9

• Automatically generate a result summary file with all results for experiment in one TSV.
• Fixed bug where printing predictions to file would cause a crash with some learners.
• Run unit tests for Python 3.3 as well as 2.7.
• More unit tests for increased coverage.

Version 0.9.8

• Fixed crash due to trying to print name of grid objective which is now a str and not a function.
• Added --version option to shell scripts.

Version 0.9.7

• Can now use any objective function scikit-learn supports for tuning (i.e.,
  any valid argument for scorer when instantiating GridSearchCV) in addition
  to those we define.
• Removed ml_metrics dependency and we now support custom weights for kappa
  (through the API only so far).
• Require's scikit-learn 0.14+.
• accuracy, quadratic_weighted_kappa, unweighted_kappa,
  f1_score_micro, and f1_score_macro functions are no longer available
  under skll.metrics. The accuracy and f1 score ones are no longer needed
  because we just use the built-in ones. As for quadratic_weighted_kappa and
  unweighted_kappa, they've been superseded by the kappa function that takes
  a weights argument.
• Fixed issue where you couldn't write prediction files if you were
  classifying using numeric classes.

Version 0.9.6

• Fixes issue with setup.py importing from package when trying to install
  it (for real this time).

Version 0.9.5

• You can now include feature files that don't have class labels in your
  featuresets. At least one feature file has to have a label though,
  because we only support supervised learning so far.
• Important: If you're using TSV files in your experiments, you should
  either name the class label column 'y' or use the new tsv_label option
  in your configuration file to specify the name of the label column. This
  was necessary to support feature files without labels.
• Fixed an issue with how version number was being imported in setup.py that
  would prevent installation if you didn't already have the prereqs
  installed on your machine.
• Made random seeds smaller to fix crash on 32-bit machines. This means that
  experiments run with previous versions of skll will yield slightly
  different results if you re-run them with v0.9.5+.
• Added megam_to_csv for converting .megam files to CSV/TSV files.
• Fixed a potential rounding problem with csv_to_megam that could slightly
  change feature values in conversion process.
• Cleaned up test_skll.py a little bit.
• Updated documentation to include missing fields that can be specified in
  config files.

Version 0.9.4

• Documentation fixes
• Added requirements.txt to manifest to fix broken PyPI release

Version 0.9.3

• Fixed bug with merging feature sets that used to cause a crash.
• If you're running scikit-learn 0.14+, use their StandardScaler, since the bug fix we include in FixedStandardScaler is in there.
• Unit tests all pass again
• Lots of little things related to using travis (which do not affect users)

Version 0.9.2

• Fixed example.cfg path issue. Updated some documentation.
• Made path in make_example_iris_data.py consistent with the updated one in example.cfg