Mlflow

MLflow

Installation

# Install MLflow
pip install mlflow

# Start the tracking UI
mlflow ui --port 5000
# Visit http://localhost:5000

Experiment Tracking

# track_experiment.py — Log parameters, metrics, and artifacts to MLflow
import mlflow
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

mlflow.set_tracking_uri("http://localhost:5000")
mlflow.set_experiment("iris-classification")

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

with mlflow.start_run(run_name="random-forest-v1"):
    # Log parameters
    n_estimators = 100
    max_depth = 5
    mlflow.log_param("n_estimators", n_estimators)
    mlflow.log_param("max_depth", max_depth)

    # Train model
    model = RandomForestClassifier(n_estimators=n_estimators, max_depth=max_depth)
    model.fit(X_train, y_train)

    # Log metrics
    accuracy = accuracy_score(y_test, model.predict(X_test))
    mlflow.log_metric("accuracy", accuracy)
    mlflow.log_metric("train_size", len(X_train))

    # Log model
    mlflow.sklearn.log_model(model, "model")

    # Log artifacts
    import json
    with open("feature_importance.json", "w") as f:
        json.dump(dict(zip(["f1","f2","f3","f4"], model.feature_importances_.tolist())), f)
    mlflow.log_artifact("feature_importance.json")

    print(f"Run ID: {mlflow.active_run().info.run_id}")
    print(f"Accuracy: {accuracy:.4f}")

Autologging

# autolog.py — Automatically log parameters, metrics, and models
import mlflow
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split

# Enable autologging for sklearn (also works with pytorch, tensorflow, xgboost, etc.)
mlflow.sklearn.autolog()

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

with mlflow.start_run():
    model = GradientBoostingClassifier(n_estimators=200, learning_rate=0.1)
    model.fit(X_train, y_train)
    # All params, metrics, and model are logged automatically

Model Registry

# model_registry.py — Register, version, and manage models
import mlflow

# Register a model from a run
model_uri = f"runs:/{run_id}/model"
model_version = mlflow.register_model(model_uri, "iris-classifier")

# Load a registered model
from mlflow.pyfunc import load_model
model = load_model("models:/iris-classifier/1")  # By version
model = load_model("models:/iris-classifier@production")  # By alias

# Set model alias
client = mlflow.tracking.MlflowClient()
client.set_registered_model_alias("iris-classifier", "production", version=1)
client.set_registered_model_alias("iris-classifier", "staging", version=2)

PyTorch Integration

# pytorch_tracking.py — Track PyTorch training with MLflow
import mlflow
import torch
import torch.nn as nn

mlflow.set_experiment("pytorch-demo")

with mlflow.start_run():
    model = nn.Sequential(nn.Linear(10, 64), nn.ReLU(), nn.Linear(64, 1))
    optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
    criterion = nn.MSELoss()

    mlflow.log_param("learning_rate", 0.001)
    mlflow.log_param("architecture", "10-64-1")

    for epoch in range(100):
        x = torch.randn(32, 10)
        y = torch.randn(32, 1)
        loss = criterion(model(x), y)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if epoch % 10 == 0:
            mlflow.log_metric("loss", loss.item(), step=epoch)

    mlflow.pytorch.log_model(model, "model")

Serving Models

# Serve a registered model as a REST API
mlflow models serve -m "models:/iris-classifier@production" -p 5001 --no-conda

# Test the endpoint
curl -X POST http://localhost:5001/invocations \
    -H "Content-Type: application/json" \
    -d '{"inputs": [[5.1, 3.5, 1.4, 0.2]]}'

Remote Tracking Server

# Start a remote tracking server with PostgreSQL backend and S3 artifact store
mlflow server \
    --backend-store-uri postgresql://user:pass@localhost:5432/mlflow \
    --default-artifact-root s3://my-mlflow-bucket/artifacts \
    --host 0.0.0.0 \
    --port 5000

Key Concepts

• Runs: Individual experiment executions with parameters, metrics, and artifacts
• Experiments: Groups of runs for comparison and organization
• Model Registry: Central hub for versioning, aliasing (staging/production), and managing models
• Autologging: One-line integration for sklearn, PyTorch, TensorFlow, XGBoost, etc.
• Artifacts: Files (models, plots, data) stored alongside runs for reproducibility
• Model serving: Deploy any registered model as a REST API with mlflow models serve