Skll

- Fix crash with GradientBoostingRegressor and MultionomialNB from typo in
previous release.
- Fix crash related to loading feature sets that contain files that are
unlabelled.
- Fix crash related to loading .megam files with unlabelled examples.

- Added new versions of kappa metrics that make it so adjacent ratings are
not penalized. For example, 1 and 2 will be considered to be equal,
whereas 1 and 3 will have a difference of 1 for when building the weights
matrix.
- Cleaned up a bit of the Sphinx documentation.
- Each module now has its own separate logger, which should make logging
messages more informative.
- Made handling of non-convertible IDs when using `ids_to_floats` uniform
across date file types. All will now raise a `ValueError` when faced
with a string that does not look like a float.
- Now raise an error when duplicate feature names are encountered in .megam
files.
- No longer set `compute_importances` for learner based on decision trees,
since that is no longer necessary as of scikit-learn 0.14.

- Added support for `DecisionTreeRegressor` and `RandomForestRegressor`.
- Fixed issue 60 with filtering examples via `cv_folds_location` file.
- Added unit tests for Grid Map mode on Travis using the scripts described
by [this gist](https://gist.github.com/dan-blanchard/6586533).
- Switched from using mix-in to decorator for handling rescaled versions of
regressors. Code's a lot simpler now.
- Added support for suppressing "Loading..." messages to most functions in
the `experiments` module.
- Refactored `convert_examples` to simply call new version of
`load_examples` that can take a list of example dictionaries in addition
to filenames.
- Made all unit tests much less verbose.
- Fix an obscure issue related to loading examples when
`SKLL_MAX_CONCURRENT_PROCESSES` is set to 1.

- Added warning when configuration files contain settings that are invalid.
- Fixed a crash because `job_results` was not defined in grid-mode.
- Cleaned up a lot of things related to unit tests and their discovery.
- Added unit tests to manifest so that people who install this through pip
could run the tests themselves if they wanted.

- Now raise an exception when using `ids_to_floats` with non-numeric IDs.
- Fixed a number of inconsistencies with `cv_folds-location` and
`ids_to_floats` (including GH issue 57).
- Fixed unit tests for `cv_folds_location` and `ids_to_floats` so that
they actually test the right things now.

- Fixed crash when using `cv_folds_location` with `ids_to_floats`.