This article introduces you to the Workflows interface and the different types of steps that can be used to create a Workflow pipeline.


Contents


The Workflows interface

To get to Workflows, click Factory > Workflows.

The Workflows screen shows all of the Workflows available on your Peak organization. From here, you can create new workflows, edit published workflows, and pause or run workflows.


Workflow steps

A Workflow pipeline starts with a trigger and is then built up by adding Workflow steps that depend on previous steps. This ensures that scripts are run in the intended order and that variables can be passed between workflow steps.

Each step can be run on a completely different image and on different resources. This means that the user can select the correct resource for each individual script. For example, the first step may require a Python base image and a GPU heavy instance, whereas the next could be a memory-intensive R script; this can be completely customised.

Workflows cannot be published until a trigger and at least one step have been defined. 

The following types of step are available:


Trigger

All workflows start with a trigger event. There are three types available:

  • Run Once trigger:
    To run the Workflow, select "Run now" once it is published.

  • Schedule trigger:
    Schedule when the Workflow runs. A basic and advanced (Cron) scheduler is available.

  • Webhook trigger:
    Trigger a Workflow to run via a webhook from another system.


Standard Step

Runs a script from a GitHub repository.


SQL Query

Runs an SQL query from either a GitHub repository or a saved query from SQL explorer.


Export Data

Exports data from a table, allowing for ordering by a specific column in ascending or descending order.


APIs

Deploys an API endpoint as part of workflow so that the results from models can be exposed to the end user.


Application Block

Configures the model training parameters for specific application blocks.

This step enables you to:

  • Specify the model / recipe that you want to train

  • Use optional hyper parameter optimization (HPO)

  • Specify the frequency at which model training takes place

  • Define hyperparameters that are used to optimize the model before training begins


Input Data

Imports input data for model training and mapping datasets to specific application blocks, for example Amazon Forecast or Amazon Personalize.