Vertex AI explain with a custom trained scikit-learn classification model

[rgulden]

Hi Google Community,

I was wondering, has anyone been able to successfully train and deploy a custom trained scikit-learn classification model and deploy it to a vertex endpoint with the feature attribution through the explain endpoint working?

Every time i define my instances, predictions and explanation_spec while uploading my model, i get errors on the endpoint for the :explain method. Specifically, i get '400 bad request' with no information on why it was a bad request.

I am using the v1beta1 ai platform python SDK and also am using a custom basic serving container. The custom container works for :predict but :explain does not work. Is there some example code out there? Is scikit-learn not supported for feature attribution? 

Thanks! Ryan

[Eduardo_Ortiz]

The easiest (and recommended) way to create a training application package uses gcloud to package and upload the application when you submit your training job.

Here you can see documentation that will guide you through all of the steps that you need to follow to implement your scikit trained model.

[rgulden · Accepted solution]

For anyone looking back on this, i was able to use the following notebook to solve my problem. It seems we need to use encoding BAG_OF_FEATURES. I am not to sure why this is required, but it seems to have done the trick for me.

https://github.com/GoogleCloudPlatform/vertex-ai-samples/blob/main/notebooks/community/ml_ops/stage4...