Apps from Models #51
Apps from Models #51
Conversation
…ture/plug-in-mlflow
…ture/plug-in-mlflow
…ture/plug-in-mlflow
…ture/plug-in-mlflow
…ture/plug-in-mlflow
…ture/plug-in-mlflow
| Place the following script in the root of your app directory and run it to | ||
| push your app to the Nextmv Cloud. This is equivalent to using the Nextmv | ||
| CLI and running `nextmv app push`. |
There was a problem hiding this comment.
I see that there is another option, but I am still missing a more CI/CD friendly way of deploying. I.e., we seem to be avoiding CLI here. If this is intentional, can we offer an entry point for nextmv-py that allows running the push without adding further scripts? Something like nextmvpy push --api-key <api-key> --app-id <app-id> maybe?
There was a problem hiding this comment.
Well, my intention for clarifying here is that people can rely on nextmv-py and not "know" about the CLI. I don’t know that the Python SDK should become a CLI tool as well, so I will defer that decision to others. In any case, that would be a follow up and I would address that in a different PR.
There was a problem hiding this comment.
Fair point. I still think this is an odd or non-idiomatic experience for a non-notebook use-case (particularly for harmonic and CI/CD reasons). It would be nice to discuss this before we change the recommendation in general. 😊
This PR introduces what I am calling the "Apps from Models" logic 👉🏻 we create a Nextmv Application from a Nextmv Python Model. This means that we can have a fully native experience in Python of creating a decision
Model, pushing it to Cloud, and running it, without ever having to leave the Python session (like a script, or a notebook).This is possible by leveraging
mlflowand their "Models from Code" logic.