Export to TensorFlow for Anomaly Detection

[pkcmn]

Hi,

API reference for IsolationForestModel contains the to_tensorflow_saved_model function (https://ydf.readthedocs.io/en/stable/py_api/IsolationForestModel/#ydf.IsolationForestModel.to_tensorflow_saved_model).
However running it raises:
ValueError: Export to TensorFlow is not yet supported for Anomaly Detection models.

I also noticed there is a TODO in the codebase mentioning that this still needs to be implemented.
Could you clarify whether TensorFlow export support for anomaly detection models is planned, or if there is a recommended workaround in the meantime?
Thanks!

[rstz]

Hi,
thank you for your interest. Our team is currently evaluating interest and impact for exposing Anomaly Detection models in Tensorflow.
If you have a significant use case, please share it with us, either publicly or via email to decision-forests-contact@google.com.

[pkcmn]

Hi, thanks for the quick follow‑up. We’re looking to use TFLite (LiteRT) to deploy anomaly‑detection models on edge devices. Our current deployment pipeline supports only TensorFlow autoencoder models, but we’d like to extend it to also support Isolation Forests. They offer strong performance while remaining lightweight, making them a good fit for edge‑resource constraints.

My idea was to export the ydf IsolationForestModel to Tensorflow and then use our existing LiteRT deployment pipeline.
Thanks!