MLflow

Server Connection and Setup

mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./artifacts --host 0.0.0.0 --port 5000
Start MLflow tracking server with SQLite backend

mlflow server --backend-store-uri postgresql://user:password@localhost/mlflow --default-artifact-root s3://bucket/path
Start MLflow tracking server with PostgreSQL backend and S3 artifact storage

export MLFLOW_TRACKING_URI=http://localhost:5000
Set MLflow tracking server URI in environment

mlflow.set_tracking_uri('http://localhost:5000')
Set MLflow tracking server URI in Python code

Experiment Management

mlflow experiments create --experiment-name
Create a new experiment

mlflow experiments list
List all experiments

mlflow experiments search --filter "name='my-experiment'"
Search for experiments by name

mlflow.set_experiment('my-experiment')
Set active experiment in Python

mlflow experiments delete --experiment-id
Delete an experiment by ID

Run Management

mlflow run
Run a project from a URI or local path

mlflow run . --experiment-name my-experiment
Run current directory project in a specific experiment

mlflow run . -P alpha=0.5
Run with a parameter

mlflow.start_run()
Start a new run in Python

mlflow.end_run()
End current run in Python

mlflow runs list --experiment-id
List runs for an experiment

mlflow runs delete -r
Delete a run by ID

Tracking

mlflow.log_param('learning_rate', 0.01)
Log a parameter in Python

mlflow.log_metric('accuracy', 0.95)
Log a metric in Python

mlflow.log_metrics({'accuracy': 0.95, 'f1': 0.90})
Log multiple metrics in Python

mlflow.log_artifact('plot.png')
Log an artifact file in Python

mlflow.log_artifacts('outputs')
Log a directory of artifacts in Python

mlflow.set_tag('version', 'v1.0')
Set a tag in Python

Model Management

mlflow.sklearn.log_model(model, 'model')
Log scikit-learn model in Python

mlflow.pytorch.log_model(model, 'model')
Log PyTorch model in Python

mlflow.tensorflow.log_model(model, 'model')
Log TensorFlow model in Python

mlflow.pyfunc.load_model('runs://model')
Load a model in Python

mlflow models serve -m runs://model -p 1234
Serve a model as REST API

mlflow models build-docker -m runs://model -n my-model
Build Docker image for a model

UI and Search

http://localhost:5000
Access MLflow UI in browser

mlflow.search_runs(experiment_ids=['0'])
Search runs for an experiment in Python

mlflow.search_runs(filter_string="metrics.accuracy > 0.9")
Search runs with metric filter in Python

mlflow.get_run()
Get a specific run in Python