Open
Description
- Install the mlflow docker containers within the system - https://mlflow.org/docs/latest/docker.html
- Get familiar with the interface and loading data for training. This will help for domain specific vocab when converting to using the REST endpoints -
- Get data from Jason for Training and understand pre-processing required of the WAV files for training (conversion to spectrograms)
- Map out using the REST API for mlflow for how we could modify and transfer data from the batai celery/django container and send it to mlflow for training
- Understand how to get the resulting model data out of mlflow and store it locally to the celery container for use in the future. This may also require creating a Django Model for model trained versions.
- Investigate getting additional artifacts out of Mlflow, for example visualizations such as confusion matrices that could be viewed through the BatAI interface
- Create a celery task that trains some model as proof-of-concept for integration with Mlflow. It should retrieve data from the database, do some training, and log the run in Mlflow with performance metrics and a manifest file
Metadata
Metadata
Assignees
Labels
No labels