Sslearn

Added
- `feature_fusion` and `probability_fusion` methods for restricted in `sslearn.restricted` module.

Fixed
- CoForest random integer is now compatible with Windows.

Added
- Add a parameter to `artificial_ssl_dataset` to force a minimum of instances. Issue 11
- Add a parameter to `artificial_ssl_dataset` to return indexes. Issue 13

Changed
- The `artificial_ssl_dataset` changed the process to generate the dataset, based in indexes. Issue 13

Fixed
- DeTriTraining now is vectorized and is faster than before.

Changed
- Hot fix for avoid problems with Pypi

Added
- Methods now support no unlabeled data. In this case, the method will return the same as the base estimator.

Changed
- In OneHotEncoder, the `sparse` parameter is now `sparse_output` to avoid a FutureWarning.

Fixed

- CoForest now is most similar to the original paper.
- TriTraining can use at least 3 n_jobs. Fixed the bug that allows using as many n_jobs as cpus in the machine.

Fixed

- Fixed a bug in TriTraining when one of the base estimators has not a random_state parameter.
- Fixed OneVsRestSSL with the random_state parameter.
- Fixed WiWTriTraining when no `instance_group` parameter is not provided.
- Fixed a FutureWarning for `sparse` parameter in `OneHotEncoder`. Changed to `sparse_output`.

Added

- CoTraining support a `threshold` parameter (default to 0.5) to control the threshold for adding new instances in the next iteration.

Fixed

- Fixed a bug in CoTraining using LabelEncoder.