Skip to main content

Celonis Product Documentation

Forward Parameters to Machine Learning Script

In order to use the full power of Machine Learning Scripts in combination with Action Flows, the forwarding of parameters is essential. That's the reason we provide you with an Action Flow Snippet that integrates desired parameter of the Action Flow into the ML tool automatically.

Follow the step-by-step guide below to implement this solution for your Action Flows snippet.

Sample Action Flow

The image below shows an Action Flow snippet which:

  1. Creates JSON data structure to forward parameters

  2. Trigger ML script with parameters

60362991.png
Configuring an Action Flow

Below you will find the step-by-step guide for configuring each module of the Action Flow above.

0. Trigger the Action Flow

Note

You can integrate any trigger sequence, e.g. Slack, Email, Celonis, etc.

An example for a Slack trigger sequence can be found at Trigger Machine Learning Script.

60362997.png
1. Customize Variables

This is the most important module in this Action Flow where we have to adjust all the data specific to your team and account like the team domain or an Application Key with the right permissions

60361485.png

Configuration:

Action Flows Module: Tools

Action: Set multiple variables

teamDomain: Enter your teamdomain, e.g. demo-action-flows

env: Enter your cluster, e.g. try, eu-1...

AppKey: e.g.:

GjV6ODBvghgv6r76r0YzkyLTkxZwjbflqjwhebfljwhebfqjhebfwlV5TEVCcjMzAHBFK0F8TXdGBTlqBWFlsVPN

notebookId: This will be added to the Template later when creating a ML App / adding Scripts to an existing ML App.

executionFileName: Name of the ML Script to be triggered, e.g. trigger_MLScript.ipynb

60361487.png
2. Pass Parameters as JSON

This module creates a JSON file that is sent as a variable in the HTML request body to be used in our ML script.

Note

The API endpoint that forwards parameters to the ML workbench script has a maximum size limit of 96 KB. This is a hard limit and cannot be changed.

This limit is typically exceeded when trying to pass files to the ML Workbench. In this case, we recommend passing the parameters of the file metadata (i.e., file id, file path, download url, etc.) and then making an API call within the ML Workbench script to retrieve the file(s).

60362703.png

Configuration:

Action Flows Module: JSON

Action: Create JSON

Parameters, e.g.:

JSON file for Generator

{
  "parameter1": "test1",
  "parameter2": "test2",
  "parameter3": "test3"
}

Possible Data Types

You can pass parameters of any type that can be handled by standard Python or libraries, such as HTML, JSON, str, int, csv-files, txt-files, etc.

60362705.png
3. Trigger ML Script

This module sends a POST Request to our ML API and triggers the previously defined ML script.

Important

You do not have to change anything in this module.

60361490.png

Configuration:

Action Flows Module: HTTP

Action: Make a Request

URL: https://{{teamDomain}}.{{env}}.celonis.cloud/machine-learning/api/executions

Method: POST

Headers:

  • Name: Authorization

  • Value: AppKey {{AppKey}}

Body type: Raw

Content type: JSON (application/json)

Request content:

{
"notebookId": "{{notebookId}}",
"executionFileName":"{{executionFileName}}",
"params": {{json}}
}
60362707.png
ML Script: Adding a ML Script to Trigger

Import the script in a new or existing ML App. The ML script will run your pre-defined Python logic and can make use of the parameters that you have forwarded.

Creating your own Logic

The script includes further documentation on how to ensure the successful execution. Make sure to follow these steps as well.

If you apply your own logic and parameters, make sure to code robustly and test the script thoroughly. If an error occurs, the backend does not give an indicator what the error was.

  1. Navigate to the Machine Learning tab.

    60361493.png

  2. Open existing App or create a new one.

    60361491.png

    Delay

    When creating a new ML App, it can take a few minutes up to one hour until it can be opened.

  3. Import the Script provided above.

    60361492.png

  4. Make sure it is uploaded successfully and displayed in the section on the left.

    60361499.png

  5. Copy the notebookId to the clipboard and add it to the variables as the value for the notebookId in the "Customize Variables" module in the Triggering Action Flow.

    60361513.png
    60361487.png

Mandatory Step for Parameters

When developing your Python code, you can define default values for the parameters to run the script by itself and test your logic.

IMPORTANT Tag the cell of your Notebook with the Tag "parameters" where you set your default values with the same data type that you will pass from the Action Flow. While developing and running the code from Jupyter, these values will be used. When you are calling the Notebook from the Action Flow, the default values will be overwritten by the ones passed by the Action Flow.

60362921.png
Downloading Action Flow blueprints

To download blueprints for Action Flows, go to Marketplace. In the Marketplace page for your Action Flow, click the Download button to get the blueprint.

Blueprints can be imported into your Celonis Platform Team so that you can quickly make the required changes without needing to build the Action Flow from scratch.

Potential Enhancements

You could integrate other apps into the Action Flow Snippet to forward parameters with data from Celonis, Google Sheets, SAP, etc.

60362713.png
Possible Use Cases

  • Run complex Python scripts in Action Flows and use the results for different use cases.

  • Integrate ML capabilities in your Action Flows.

Potential Alternatives

You could replace the triggering Slack module with a Scheduling to run the ML Script frequently.

You could also replace the Slack module at the end with a messaging module of your choice such as Email or Microsoft Teams.