Workaround for scikit-learn 1.6.0 Incompatibility in IRIS 2024.3 AutoML

Introduction

In InterSystems IRIS 2024.3 and subsequent IRIS versions, the AutoML component is now delivered as a separate Python package that is installed after installation. Unfortunately, some recent versions of Python packages that AutoML relies on have introduced incompatibilities, and can cause failures when training models (TRAIN MODEL statement). If you see an error mentioning "TypeError" and the keyword argument "fit_params" or "sklearn_tags", read on for a quick fix.

Root Cause

• scikit-learn updated to version 1.6.0, deprecating fit_params.
• scikeras is no longer updating the "wrappers" for Tensorflow.
• Installing these packages in IRIS 2024.3 without pinned dependencies leads to training errors.

Symptoms

You may encounter:

<TypeError>: got an unexpected keyword argument 'fit_params'

or

[%msg: <ERROR #5002: ObjectScript error: <PYTHON EXCEPTION> *<class 'AttributeError'>: 'super' object has no attribute '__sklearn_tags__' [^%ML.AutoML.Provider.1:%SYS]>]

during the model training phase.

Workaround: 

1. Uninstall Existing Packages (if needed)

pip uninstall scikit-learn xgboost keras ...

2. Install via Snapshot

• download requirementsSnapshot.txt (right-click "Save page as...")
• install automl:

pip install --index-url https://registry.intersystems.com/pypi/simple \
  --no-cache-dir \
  --target /path/to/iris/python \
  -r /path/to/requirementsSnapshot.txt

 3. Verify

• Run your smoke tests to confirm no fit_params errors.

• Ensure scikit-learn is pinned at <1.6.0.

Note: The exact means by which you can obtain or access requirementsSnapshot.txt are still under discussion. For now, please contact your InterSystems representative or consult the official documentation channels for the most up-to-date link or distribution process.

Alternative: Pin scikit-learn <1.6.0

If you prefer not to use the snapshot file, you can manually pin the version:

pip install scikit-learn<1.6.0 xgboost keras ...

This will work around the issue until XGBoost and Keras officially update their packages.

Next Steps

• InterSystems has implemented changes to the AutoML code to solve the issue in future releases -- this is mostly an issue for older containers such as 2024.3, the first release to use the post-installation procedure to install IntegratedML.
• InterSystems is issuing a customer alert for the IRIS 2024.3 ml containers (containers.intersystems.com/intersystems/iris-ml:2024.3), that references this article.
• If you have questions or run into issues, feel free to comment below or reach out to your InterSystems support contact.

Additional Resources:

• InterSystems IntegratedML AutoML installation documentation
• scikit-learn 1.6.0 release notes 
• XGBoost compatibility discussion on github.com