Creating a training pipeline

Introduction

Here we discuss running the training pipeline for on-cloud implementation. On-premise implementation is almost identical, and the difference is highlighted for each of the components. In order to create a training pipeline, click on the “create training pipeline” button, and you will proceed to the first step – “Pipeline details.”

Pipeline details

In this section, the following fields are filled by the user. Please note that none of them are mandatory. The name of the pipeline and the description would be displayed on the “pipeline runs” page. The git username and git token give access to your code to Robotika. You can skip the git credentials and directly upload the files when configuring the components. The organization name is fixed and cannot be modified.

Pipeline component config

For the pipeline to run the following components should be configured:

1. ExampleGen
2. Transform
3. Trainer
4. Evaluator
5. InfraValidator
6. Pusher

The components that are not configured on this step but contains valuable information during the debugging and after the deployment are:

1. StatisticGen
2. SchemaGen
3. ExampleValidator
4. Tuner

Please refer to the page corresponding to the specific component for more details.

Scheduler

The pipeline Scheduler is responsible for the retraining of the ML pipeline. Currently, Robotika supports only retraining based on the time schedule. Please note the first training run occurs as soon as a user clicks the button “Save.”