Support `experiment_config` in HyperparameterTuner for unified lineage across training and tuning phases

[cfregly]

Is there a way to pass experiment_config to HyperparameterTuner to have unified lineage tracking across all phases of the model?

What is the recommended way to track end-to-end lineage across training and tuning phases? Should we just manually create a Tracker and log the hyperparameters, objective metrics, and training result metrics?

I'd like to see symmetry between the Estimator and HyperparameterTuner APIs - and pass experiment_config to both.

Any guidance would be appreciated.

[chuyang-deng]

Hi @cfregly. Thanks for using SageMaker! Unfortunately Hyperparameter Tuning currently does not support ExperimentConfig in the service API yet.

However, once we got an update on the service side, we will update SageMaker Python SDK as well.

[cfregly]

Any update?

[leninkumar-sv-tiger]

Any update on this ?

[krishnashanker-amt]

Would love it if this could be implemented.

[paulm-1]

Same

[pdifranc]

Would be really useful to have it implemented rather then manually associate the training jobs after the HyperparameterTuner completes its job