INTRODUCTION:

Datapred’s modeling engine for continuous intelligence is the most powerful and flexible way to apply machine learning to time series (or streaming data, or event streams).

Datapred streamlines everything that is specific about time series modeling at every stage of the ML pipeline, ensuring significant performance gains and a 10x faster garage-to-factory cycle.

NOTABLE FEATURES

Preprocessing

- Prevention of "future leakage" (automatic data alignment)
- Target stationarization (seasonality, outliers)

Modeling

- Adaptation of batch algos to the sequential context (give us a fit/predict, we do the rest)
- Online aggregation of multiple algos
- Custom loss functions

Post-processing

- Feature re-integration (seasonality, outliers)
- Online integration of operational constraints

PRODUCTION-READY

Datapred accelerates the transition from good machine learning model to production-ready machine learning application:

- It packages over 30 DevOps open-source components to help you kick-start, deploy and maintain your applications.
- It uses a graph structure that facilitates collaborative development, parallel vs distributed computations, continuous improvement and component recycling.
- It includes built-in connectors to standard databases, connector templates for non-standard databases, and is natively Linux/Docker-compatible.

Use Datapred from a Jupyter Notebook, with any ML library.

Datapred for Azure includes a full documentation and detailed tutorials on most time series challenges (e.g. sales forecasts, anomaly detection) that you can use as quick starting points for your projects.