Pipeline introduction
The ML pipeline is a way to automate the ML process, including (but not limited to) data preprocessing, training, inference, and monitoring of the ML model. In Robotika, ML pipelines are component-based and can be managed by intuitive UX. The Robotika components are inherited from the TFx components, so technically, they are upgraded versions of the original TFx components with an easy-to-use UX interface. Additionally, we introduced a few new components. Please refer to the component documentation page for the details regarding each of the components.
Training vs Inference pipeline
In Robotika, users can create training and inference pipelines. The training pipeline is responsible for ML model training. Additionally, it creates an endpoint for the on-demand prediction. The Inference pipeline is responsible for batch inference prediction. In other words, with an inference pipeline, you can adapt and schedule prediction jobs for large batches of data.
On-cloud and on-premise run
Robotika enables running the pipelines either on-cloud or on-premise. By running on-cloud, all the data and pipeline metadata are stored, and the computation is performed on our servers. By running on-premise, all the computation will be performed in your data center. We care about data privacy, and the data will not leave your servers. However, the metadata for running the pipelines is stored on our servers. This approach is in compliance with GDPR and does not violate the data privacy requirements, even for such strict industries as Banking and Health care.