mlflow

A minimal, Wolfi-based image for MLflow, an open source platform for the machine learning lifecycle.

Download this Image

The image is available on cgr.dev:

docker pull cgr.dev/chainguard/mlflow:latest

MLflow Usage

MLflow's default entrypoint is Python, enabling us to run experiments directly:

docker run -it cgr.dev/chainguard/mlflow:latest <your experiment>.py

Otherwise, we can override the entrypoint and interact with MLflow:

docker run -it --entrypoint mlflow cgr.dev/chainguard/mlflow:latest <options>

MLflow Tracking Usage

MLflow provides a UI, MLflow Tracking, that allows the user to track 'runs' (the execution of data science code) via visualizations of metrics, parameters, and artifacts.

To start the UI, open a terminal and run:

docker run -it -p 5000:5000 --entrypoint mlflow cgr.dev/chainguard/mlflow:latest ui

While the UI defaults to running on port 5000, you can use a different port via passing -p <PORT> as a command line option. Ensure Docker also maps to the correct port.

You should now be able to access the UI at localhost:5000.

The Tracking API can now be leveraged to record metrics, parameters, and artifacts:

import mlflow

# Set the MLflow tracking URI
mlflow.set_tracking_uri("http://localhost:5000")

# Start an experiment
mlflow.set_experiment("my_experiment")

with mlflow.start_run():
    # Log parameters, metrics, and artifacts
    mlflow.log_param("param1", value1)
    mlflow.log_metric("metric1", value2)
    mlflow.log_artifact("path/to/artifact")
    # Train and log model
    mlflow.sklearn.log_model(model, "model")

Ensure that the tracking URI correctly reflects where the MLflow server is running.

For additional documentation covering MLflow Tracking, see the official docs.