Pipeline as model

Pipeline as model

• Introduction to mlflow models
  • What are Mlflow Models ?
  • Pre-requisite for converting a pipeline to a mlflow model
• Scikit-learn like Kedro pipelines - Automatically log the inference pipeline after training
  • Getting started with pipeline_ml_factory
  • Advanced configuration for pipeline_ml_factory
    • Register the model as a new version in the mlflow registry
  • Complete step by step demo project with code
• Deployment patterns for kedro pipelines as model
  • Deploying a KedroPipelineModel
  • Frequently asked questions
• Custom registering of a KedroPipelineModel
  • Log a pipeline to mlflow programatically with KedroPipelineModel custom mlflow model
  • Log a pipeline to mlflow with the CLI

kedro-mlflow as a mlops framework

• Why we need a mlops framework to manage machine learning development lifecycle
  • Machine learning deployment is hard because it comes with a lot of constraints and no adequate tooling
    • Identifying the challenges to address when deploying machine learning
    • A comparison between traditional software development and machine learning projects
      • ML and traditional software have different moving parts
      • ML and traditional software have different development lifecycles
  • Deployment issues addressed by kedro-mlflow and their solutions
    • Out of scope
    • Issue 1: The training process is poorly reproducible
    • Issue 2: The data scientist and stakeholders focus on training
    • Issue 3: Inference and training are entirely decoupled
    • Issue 4: Data scientists do not handle business objects
    • Overcoming these problems: support an organisational solution with an efficient tool
• The components of a machine learning application
  • Definition: apps of a machine learning projects
  • Difference between an app and a Kedro pipeline
  • Apps development lifecycle in a machine learning project
    • The data scientist creates at least part of the 3 apps
    • The etl_app
    • The ml_app
    • The user_app
• kedro-mlflow mlops solution
  • Reminder
  • Enforcing these principles with a dedicated tool
    • Synchronizing training and inference pipeline
    • Packaging and serving a Kedro Pipeline
    • kedro-mlflow’s magic: inference autologging
    • Reuse the model in kedro