MLFlow tutorials

Some examples of how to use MLFlow. Those examples were used in my presentation at 2020's Python Brasil conference.

MLFlow

MLflow is a platform to manage the ML lifecycle, including experimentation, reproducibility, deployment, and a central model registry. MLflow currently offers four components:

• Tracking:

Record experiment configs: code, data, parameters, metrics. MLfow provides a UI to access those configs and compare many experiments.

• Projects

Package data science code in a format to reproduce experiments on any platform

• Model

Package machine learning models to be deployed in diverse serving environments. It's possible to serve models as APIs with only 1 command-line

• Model Registry

A central repository of models, with versioning.

Tracking

The folder xgboost has an example of how to use MLFlow Tracking to log the parameters, metrics, and models of an experiment using XGBoost. To run this example:

cd xgboost
python train.py

To see the results in the MLFlow UI dashboard:

mlflow ui

Project

To run the xgboost example as Project (inside a Conda Environment):

 mlflow run xgboost -P colsample-bytree=0.7

To run the xgboost example as Project (inside a Docker):

docker build -t mlflow-iris-xgboost -f xgboost_docker/Dockerfile xgboost_docker/

mlflow run xgboost_docker -P colsample-bytree=0.7

Model

To serve a trained model as an API:

mlflow models serve -m runs:/<RUN-ID>/<MODEL-PATH>

Accesing the API:

curl http://127.0.0.1:5001/invocations -H 'Content-Type: application/json' -d '{         
	"columns": ["0", "1", "2", "3"], 
	"data": [ [5.1, 3.5, 1.4, 0.2], [5.9, 3. , 5.1, 1.8] ]     
	}'


[[0.95, 0.02, 0.02], [0.03, 0.12, 0.83]]

Registry Model

To run a server to track the experiment configs (on a database, mandatory to a Registry Model Server), save the artifacts (on an S3 bucket) and register the models:

mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root s3://<bucket>/<path>/ --host 0.0.0.0

To run your experiment in your machine and use the server to save the experiment configs and the artifacts, you need to define the MLFLOW_TRACKING_URI environment variable. You need the permissions to access the server and/or the S3 bucket.

export MLFLOW_TRACKING_URI='http://127.0.0.1:50000' 
mlflow run xgboost

To serve a registered trained model of the Production stage as an API. You can register a model and put it in the Production stage using the UI or via code.

export MLFLOW_TRACKING_URI=http://127.0.0.1:50000 
mlflow models serve -m "models:/Iris/Production"

Accesing the API:

curl http://127.0.0.1:5001/invocations -H 'Content-Type: application/json' -d '{         
	"columns": ["0", "1", "2", "3"], 
	"data": [ [5.1, 3.5, 1.4, 0.2], [5.9, 3. , 5.1, 1.8] ]     
	}'


[[0.95, 0.02, 0.02], [0.03, 0.12, 0.83]]