How long have you used MLflow for what kind of models? Are you using it personally or do you use it as a team?
As far as I can tell, it cannot deploy models that use high dimensional input. It can log these models, and it can deploy them, but the "deployed" model can only accept two dimensional data.
This is an open issue[0] with open pull requests. We're experimenting with MLflow for our platform (https://iko.ai), but we automatically detect models, metrics, and parameters and then track that with MLflow instead of haivng people pollute their notebook code with MLflow code. We don't like leaking that detail (tracking code) into the notebook.
We circumvented these issues (https://iko.ai/docs/appbook/#deploying-a-model). The models can either be deployed, or you can build a Docker image and push it to a registry and do whatever you want with it (run a container and invoke the model's endpoint, for example).
As far as I can tell, it cannot deploy models that use high dimensional input. It can log these models, and it can deploy them, but the "deployed" model can only accept two dimensional data.
This is an open issue[0] with open pull requests. We're experimenting with MLflow for our platform (https://iko.ai), but we automatically detect models, metrics, and parameters and then track that with MLflow instead of haivng people pollute their notebook code with MLflow code. We don't like leaking that detail (tracking code) into the notebook.
We circumvented these issues (https://iko.ai/docs/appbook/#deploying-a-model). The models can either be deployed, or you can build a Docker image and push it to a registry and do whatever you want with it (run a container and invoke the model's endpoint, for example).
- [0]: https://github.com/mlflow/mlflow/issues/3570