Palladium

=====================

- Added MANIFEST.in to include VERSION file to distribution

===================

- Added multiple confuration file support, i.e., PALLADIUM_CONFIG can
now point to a list of configuration files where the configuration
dict is updated successively

- Added functionality to handle multiple predict service entry points

- Added support for nested lists in configuration

- Added `here` variable to be used in configuration pointing to the
parent path of the current configuration file

- Added support for PluggableDecorators to DatasetLoaders
('load_data_decorators' in config)

- Added module to be used with WSGI servers initializing configuration
without delay

- Added parameter for poolclass in sqlalchemy create_engine call in
Database model persister. Default poolclass is set to None now to
avoid DB connections being open for longer time periods

- Updated requirements in order to use newer versions of dependencies

- Added Python 3.6 to Travis CI test settings

- Changed deprecated sklearn.grid_search.GridSearchCV to current
sklearn.model_selection.GridSearchCV

===================

- Updated requirements in order to use newer versions of dependencies

- Added information about virtual memory size (vms) to /alive entry
point

- Added Python 3.5 to and removed Python 3.3 from Travis CI test
settings

=================

- Added a streaming server (`pld-stream`) which listens to stdin,
processes line by line, and returns predictions.

- Added support for batch processing through the web API. You can now
use a POST request to predict for a number of samples at once.
(Please see section "Upgrading" in the docs for details about
changes to the PredictService API.)

- Added support for selecting active model used for predictions.
Previously, the latest trained model was used. (Please see section
"Upgrading" in the docs for details on how to upgrade model
persisters (database or file) to the latest version.

- Added a new command `pld-admin` to activate and delete models.

- Added a new command `pld-dockerize` to create application specific
docker images for Palladium-based services. Previously, this used
to be handled by a script in `addons/docker/palladium_app_image`.

- Added a ModelPersister for RDBMs that don't have BLOB.

- Documented scripts.

===================

- Added initial version of Palladium
- Added scripts to fit, grid-search, test, and list models
- Added script to provide model as web service
- Added scripts to create Docker images of Palladium-based services
- Added documentation and tutorial
- Added Iris classification as sample application
- Added examples how to integrate R and Julia models
- Added information on how to setup Palladium with Mesos / Marathon