MLflow Integration

MLflow is a free plugin for nself that adds ML experiment tracking, model versioning, and a model registry to your backend stack. As of v1.0.9, MLflow is no longer a built-in optional service — it installs via the plugin system.

Installation

# Install the MLflow plugin (free, no license required)
nself plugin install mlflow

# Rebuild and start to apply
nself build && nself start

Once installed, MLflow is accessible at:

• Development: https://mlflow.local.nself.org
• Production: https://mlflow.<your-domain>

Prerequisites

MLflow requires:

• PostgreSQL — for backend store (already included as required service)
• MinIO — for artifact storage (enable with MINIO_ENABLED=true)

MLflow Commands

After the plugin is installed, the following commands are available:

Managing Experiments

# List all experiments
nself mlflow experiments list

# Create a new experiment
nself mlflow experiments create "My Experiment"

# Delete an experiment by ID
nself mlflow experiments delete 1

# List runs from experiment ID 1
nself mlflow runs 1

Python Integration

Log experiments from your Python ML code:

import mlflow

# Set tracking URI to your nself MLflow instance
mlflow.set_tracking_uri("http://mlflow.local.nself.org")

# Create or set experiment
mlflow.set_experiment("text-classification")

# Start a run
with mlflow.start_run():
    mlflow.log_param("learning_rate", 0.01)
    mlflow.log_param("epochs", 100)

    model = train_model(learning_rate=0.01, epochs=100)

    mlflow.log_metric("accuracy", 0.95)
    mlflow.log_metric("f1_score", 0.93)
    mlflow.sklearn.log_model(model, "model")
    mlflow.log_artifact("confusion_matrix.png")

Model Registry

import mlflow

# Register a model
model_uri = "runs:/<run_id>/model"
mlflow.register_model(model_uri, "TextClassifier")

# Load a registered model
model = mlflow.pyfunc.load_model("models:/TextClassifier/Production")

# Transition model stage
client = mlflow.tracking.MlflowClient()
client.transition_model_version_stage(
    name="TextClassifier",
    version=1,
    stage="Production"
)

Configuration

Environment variables set automatically by the plugin after installation:

MLFLOW_VERSION=2.9.2
MLFLOW_PORT=5000
MLFLOW_USERNAME=admin
MLFLOW_PASSWORD=<auto-generated>
MLFLOW_DB_NAME=mlflow
MLFLOW_ARTIFACTS_BUCKET=mlflow-artifacts

Configure Settings

# Set username
nself mlflow configure username admin

# Set/generate password
nself mlflow configure password

# Change port
nself mlflow configure port 5001

Troubleshooting