An open source platform for the machine learning lifecycle mlflow.org/

context

  • Provide developers with similar benefits on platforms such as Google TFX and Facebook FBLearner Flow
  • Any tool and algorithm can be supported

The project architecture

MLflow Tracking

  • Record and query experiments: code, data, configuration, and results
  • www.mlflow.org/docs/latest…

MLflow Projects

  • A packaging format that can be run repeatedly on any platform
  • www.mlflow.org/docs/latest…

MLflow Models

  • A common format for sending models to various deployment tools
  • www.mlflow.org/docs/latest…

Model Registry

  • Central repository: Store, annotate, discover, and manage models
  • Mlflow.org/docs/latest…

Projects using

  • Anaconda environment installation, easy access to packages and packages can be managed, while the environment can be centrally managed distribution

Mlflow installation

pip install mlflow
OR
conda install mlflow
Copy the code

The UI to start

  • examples/mlflow_tracking.py
cd examples
mlflow ui
Copy the code

Model run

Python3 Sklearn_logistic_regression /train.py # API start MLflow models serve -m runs:/<Model_id>/model --port 1234 May be an error: mlflow utils. Process. ShellCommandException: Non - zero exitcode: 1 mlflow models serve -m runs:/<Model_id>/model --port 1234 --no-condaCopy the code

Stream processing Demo

  • Github.com/mlflow/mlfl…

thinking

  • MLflow Monitoring
  • MLFlow is still an auxiliary tool similar to Pipeline at this stage. It defines Pipeline tools and standards and lacks the concept of AI platform
  • How to integrate with Spark and Docker

That’s all!