Page MenuHomePhabricator

Testing Model
Closed, ResolvedPublic

Description

The end goal for the proof of concept sprint is to serve a simple model over KFServing. To do that we need to create a simple model to serve.

I have no requirements for the model other than it be tree based and containerized per KF's guidelines.

Related Objects

StatusSubtypeAssignedTask
ResolvedNone
Resolved ACraze

Event Timeline

Here's a simple BoostedTreesClassifier that I trained on the Titanic dataset today. It's in the Tensorflow SavedModel format, which is required by KFServing/TFServing:

Code: https://gist.github.com/accraze/8fd30ba56bfd688ff9e5506976b86901

I plan to do a similar model in sklearn next week so we can test the multi-framework support.

We'll need to load the model binary from storage using something like pvc: https://github.com/kubeflow/kfserving/tree/master/docs/samples/storage/pvc

Great, it all looks good. I am going to close this task