hello_world

mlflow-fun - Hello World

Simple Hello World that demonstrates the different ways to run an MLflow experiment.

For details see MLflow documentation - Running Projects.

Synopsis of hello_world.py:

• Creates an experiment HelloWorld if it does not exist. You can optionally override with the standard MLFLOW_EXPERIMENT_NAME environment variable.
• Logs parameters, metrics and tags.
• No ML training.
• Optionally writes an artifact.

The different ways to run an experiment:

• Unmanaged without mlflow
  • Command-line python
  • Jupyter notebook
• Using mlflow run with MLproject
  • mlflow run local
  • mlflow run git
  • mlflow run remote

Setup

External tracking server

export MLFLOW_TRACKING_URI=http://localhost:5000

Databricks managed tracking server

export MLFLOW_TRACKING_URI=databricks

The token and tracking server URL will be picked up from your Databricks CLI ~/.databrickscfg default profile.

Running

Unmanaged without mlflow run

Command-line python

python hello_world.py

Jupyter notebook

See hello_world.ipynb.

export MLFLOW_TRACKING_URI=http://localhost:5000
jupyter notebook

Using mlflow run

mlflow run local

mlflow run . -Palpha=.01 -Prun_origin=LocalRun -Plog_artifact=True

You can also specify an experiment ID:

mlflow run . --experiment-id=2019 -Palpha=.01 -Prun_origin=LocalRun -Plog_artifact=True

mlflow run git

mlflow run  https://github.com/amesar/mlflow-fun.git#examples/hello_world \
  --experiment-id=2019 \
  -Palpha=100 -Prun_origin=GitRun -Plog_artifact=True

mlflow run Databricks remote

Run against Databricks. See Remote Execution on Databricks and cluster.json.

mlflow run  https://github.com/amesar/mlflow-fun.git#examples/hello_world \
  --experiment-id=2019 \
  -Palpha=100 -Prun_origin=RemoteRun -Plog_artifact=True \
  -m databricks --cluster-spec cluster.json