MLflow Plugin

Run MLflow as a managed service inside your nSelf stack. Track ML experiments, log metrics and artifacts, compare model runs, and register production models — all self-hosted on your own infrastructure.

Overview

• Experiment Tracking — log parameters, metrics, and tags per run
• Artifact Store — store model binaries and datasets in MinIO (when enabled) or local disk
• Model Registry — version, stage (Staging / Production), and annotate models
• MLflow UI — full web UI available at a dedicated port
• PostgreSQL Backend — uses nSelf's managed Postgres for run metadata

Installation

nself plugin install mlflow

nself plugin status mlflow

Configuration

# Port for MLflow UI (default: 5000)
MLFLOW_PORT=5000

# Artifact storage (uses MinIO if MINIO_URL is set, otherwise local)
MLFLOW_ARTIFACT_ROOT=s3://mlflow-artifacts
# or
MLFLOW_ARTIFACT_ROOT=./data/mlflow/artifacts

# PostgreSQL backend (inherits DATABASE_URL by default)
MLFLOW_DATABASE_URL=postgresql://user:pass@localhost:5432/nself

Quick Start

import mlflow

mlflow.set_tracking_uri("http://your-nself-host:5000")
mlflow.set_experiment("my-experiment")

with mlflow.start_run():
    mlflow.log_param("learning_rate", 0.01)
    mlflow.log_metric("accuracy", 0.94)
    mlflow.sklearn.log_model(model, "model")