We are excited to announce PredictionIO version 0.9 is now available. We want to thank you for all your feedback and contributions so far!
These are the major improvements in 0.9:
Machine Learning Analytics
Using the Event Server, our open source machine learning analytics layer for
unifying events from multiple platforms, we now support batch
of events using the
pio import command.
Furthermore, you can now export data to Apache Parquet format with the
export command. We already support several popular analytics
including IPython Notebook, Tableau and Zeppelin.
Engine Templates allow you to quickly build production-ready predictive **engines or customize templates for your unique prediction requirements.
We recently announced our first 5 templates for E-commerce use cases:
- E-Commerce Recommendation
- Complementary Purchase
- Product Ranking
- Similar Product
- Lead Scoring
Engine Templates follow PredictionIO’s DASE architectural pattern (Data, Algorithm, Serving and Evaluation) allowing you to re-use and customize individual components. This separation of concern shortens the development cycle and enable rapid deployment.
pio template list command you can now see the available Engine
Evaluation Metrics & Hyperparameter Tuning
We now support Evaluation Metrics and Hyperparameter
the “E” in DASE, for example, for
our Engine Template supports
as a simple metric to compare the predicted result returned from the Engine and
the actual result which we obtained from the test data. The
pio eval command
kick starts the evaluation and we support implementing and contributing you own
Engine Debugging & Troubleshooting
To further support development of predictive engines we now offer methods for debugging, including partial engine training and sanity checking, see our Troubleshooting Engine Development docs.
We are delighted by the support from the community and we look forward to hearing your feedback on this latest release!