OpenMetadata/docs/connectors/mlflow.md

description
This guide will help install the MLflow connector and run it manually

MLflow

{% hint style="info" %} Prerequisites

OpenMetadata is built using Java, DropWizard, Jetty, and MySQL.

1. Python 3.7 or above {% endhint %}

Install from PyPI

{% tabs %} {% tab title="Install Using PyPI" %}

pip install 'openmetadata-ingestion[mlflow]'

{% endtab %} {% endtabs %}

Run Manually

metadata ingest -c ./examples/workflows/mlflow.json

Configuration

{% code title="mlflow.json" %}

{
  "source": {
    "type": "mlflow",
    "config": {
      "tracking_uri": "http://localhost:5000",
      "registry_uri": "mysql+pymysql://mlflow:password@localhost:3307/experiments"
    }
 ...

{% endcode %}

1. tracking_uri - MLflow server containing the tracking information of runs and experiments (docs).
2. registry_uri - Backend store where the Tracking Server stores experiment and run metadata (docs).

Publish to OpenMetadata

Below is the configuration to publish MLflow data into the OpenMetadata service.

Add optionally pii processor and metadata-rest sink along with metadata-server config

{% code title="mlflow.json" %}

{
  "source": {
    "type": "mlflow",
    "config": {
      "tracking_uri": "http://localhost:5000",
      "registry_uri": "mysql+pymysql://mlflow:password@localhost:3307/experiments"
    }
  },
  "sink": {
    "type": "metadata-rest",
    "config": {}
  },
  "metadata_server": {
    "type": "metadata-server",
    "config": {
      "api_endpoint": "http://localhost:8585/api",
      "auth_provider_type": "no-auth"
    }
  }
}

{% endcode %}