Machine Learning API

[m-mohr]

The API shall provide core functionalities and operators for training, parameterization, prediction and validation of Machine Learning models. (req. 48)

Could be an endpoint /models for pre-trained models and mostly all other things are hopefully possible with dedicated processes?

[jdries]

We will start experimenting with the training of a random forest model, this model can then indeed be used for inference. A fully generic ML API is out of scope.
So something like:
ml_features.train_model(type='randomforest').execute_batch(out_format="ml_model_format")

ml_features will need to be some kind of datacube containing features and their labels. Not sure if this can still be a regular datacube.

[m-mohr]

A fully generic ML API is out of scope.

Just to make this clear to externals: It is out of scope for the openEO Platform project, not necessarily out of scope for the openEO API.

ml_features will need to be some kind of datacube containing features and their labels. Not sure if this can still be a regular datacube.

I'm not so much into the ML terminology. To clarify: Features as in vector features? Which would lead us to vector data cubes?