Quickstart

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

• MLflow Tracking (experiment tracking)

• MLflow Projects (code packaging format for reproducible runs using Conda on Data Science Jobs and Data Flow)

• MLflow Models (package models for deployment in real time scoring, and batch scoring)

• Model Registry (manage models)

Using MLflow with Oracle Cloud Infrastructure (OCI) Data Science you will first need to install the Oracle OCI MLflow plugin:

Note

The OCI MLflow plugin will also install (if necessary) the mlflow and oracle-ads packages

Package Name	Latest Version
MLflow
oracle-ads

• Install the oci-mlflow plugin

  pip install oci-mlflow
  

• Test oci-mlflow plugin setup

  mlflow deployments help -t oci-datascience
  

Background reading to understand the concepts of MLflow and OCI Data Science:

• Getting started with OCI Data Science Jobs

• Getting started with Oracle Accelerated Data Science SDK to simplify creating and running Jobs

• Getting started with Data Science Environments

• Getting started with Custom Conda Environments

Authentication and Policies:

• Getting started with OCI Data Science Policies

• API Key-Based Authentication - api_key

• Resource Principal Authentication - resource_principal

• Instance Principal Authentication - instance_principal

OCI Integration Points

The oci_mlflow plugin enables OCI users to use OCI resources to manage their machine learning usecase life cycle. This table below provides the mapping between the MLflow features and the OCI resources that are used.

Note

MLflow Use Case	OCI Resource
User running machine learning experiments on notebook, logs model artifacts, model performance etc	Data Science Jobs, Object Storage, MySQL
Batch workloads using spark	Data Flow, Object Storage, MySQL
Model Deployment	Data Science Model Deployment
User running machine learning experiments on notebook, logs model artifacts, model performance etc	Object Storage, MySQL