Datapipeline issue between Celonis Studio and Machine LEarnign Workbench

This is a  issue that usually stems from a conflict between legacy packages (celonis_ml or older pycelonis 1.x versions) and updated Python/scikit-learn ecosystems. Because older Machine Learning Workbench (MLWB) environments cannot be upgraded in-place, the underlying dependencies eventually break.

Short-term workarounds and the recommended long-term fix:

If you need to patch your current notebook immediately, try the following in your MLWB terminal:

• Fix the sklearn dependency naming block: Older installers explicitly look for the deprecated sklearn package name instead of scikit-learn. Try explicitly installing scikit-learn first, then installing your package:

  Bash

   

  pip install scikit-learn
  

• Target a stable legacy version: If you are hitting InvalidVersion or signature errors, completely clean out and target a stable 1.x release of PyCelonis:

  Bash

   

  pip uninstall pycelonis -y
  pip install pycelonis==1.7.6  # or 1.7.7+ depending on your specific error
  

2. The Recommended Long-Term Fix (Environment Refresh)

Since Celonis does not support in-place upgrades for older MLWB underlying instances, the cleanest path forward is to migrate to a fresh environment and modern syntax:

1. Backup your work: Commit your current notebooks/code to Git or export them as a ZIP file.

2. Provision a new MLWB: Create a brand-new Workbench instance from the Celonis Studio overview page. Newly provisioned workbenches automatically inherit the latest Python environment and updated core dependencies.

3. Migrate to Modern PyCelonis: Move away from the legacy celonis_ml wrapper and use the native pycelonis API to pull data models directly. For example:

    

    

Hey ​@Sagar_MTB , 

Current version is 2.x (the 1.x line is deprecated). If you hit dependency issues during the upgrade, the cleanest path is actually to provision a fresh MLWB environment from Celonis Studio — newly provisioned workbenches come with the latest pycelonis and all dependencies pre-installed and pre-configured.

Docs for reference:
https://celonis.github.io/pycelonis/latest/
https://docs.celonis.com/en/getting-started-with-the-ml-workbench.html

Hi, Thank you. I finally resolved the ML Workbench issue by creating a new Workbench instance, and now the connection is successful. Appreciate your help!
However, I'm facing a new problem. PyCelonis connects successfully using get_celonis(), and I can access objects like celonis.data_integration, celonis.studio, and celonis.apps. But when I run celonis.data_integration.get_data_pools(), it returns "Number of pools: 0". Similarly, celonis.studio.get_spaces() does not return any spaces. The logs also show permissions as empty lists (e.g., ml-workbench permissions: [], team permissions: []). My goal is to access a Data Model/Knowledge Model containing a table from ML Workbench. Has anyone experienced a situation where ML Workbench authenticates successfully but cannot see any Data Pools or Studio Spaces? Does this indicate a permissions issue or does the Workbench need to be explicitly connected to a Data Pool/Space? Any guidance would be greatly appreciated. Thanks!

check the following