Open withsmilo opened 5 years ago
I think it is a good idea. I always thought it'd be better if clipper can be integrated with other ml maintenance tools. It can be also a good advertisement of our platform. One concern is if we have enough bandwidth to conduct a project. After discussing what our next release includes, we should probably find a way to look for more contributors.
After sketching first look, I will share it with you through this issue, and ask you to review it.
@withsmilo Did you have the time yet to make some initial sketch?
@T-800Model101 , I'm so sorry about late reply. this issue has been postponed because I had to resolve urgent issues in my project.
MLflow(https://github.com/mlflow/mlflow) offers a set of simple APIs that can be used for model tracking and pipeline maintenance, and so on. Also, it deploys easily a ML model to the platforms such as Azure ML and AWS SageMaker. It seems to provide some plug-in to connect with them.
What do you think of integrating Clipper with MLflow? MLflow creates a pickle file contained with ML model in the same way as Clipper. If we create a Clipper Dockerfile with this pickle file, we will be able to serve it in the Clipper.
This is just idea. I'd like to listen to other opinion.