MLOps Quick Reference Guide

Quick reference for MLOps tools and practices.

Table of Contents

• Version Control
• Experiment Tracking
• Model Registry
• CI/CD
• Common Issues & Solutions
• Best Practices Checklist

Version Control

DVC Commands

# Initialize
dvc init

# Track files
dvc add data/train.csv

# Create pipeline
dvc run -n step -d input -o output python script.py

# Reproduce
dvc repro

Git LFS

# Install
git lfs install

# Track
git lfs track "*.pkl"

Experiment Tracking

MLflow

import mlflow

mlflow.set_experiment("experiment")
with mlflow.start_run():
    mlflow.log_param("param", value)
    mlflow.log_metric("metric", value)
    mlflow.sklearn.log_model(model, "model")

Weights & Biases

import wandb
wandb.init(project="project")
wandb.log({"metric": value})

Model Registry

# Register
mlflow.register_model("runs:/<id>/model", "ModelName")

# Transition
client.transition_model_version_stage("ModelName", 1, "Production")

CI/CD

GitHub Actions

- name: Run tests
  run: pytest

- name: Train model
  run: python train.py

Common Issues & Solutions

Issue 1: Large Files in Git

Solution: Use DVC or Git LFS

Issue 2: Lost Experiments

Solution: Use MLflow from the start

Best Practices Checklist

• Version control code, data, models
• Track all experiments
• Use model registry
• Set up CI/CD
• Document everything
• Monitor models in production

Try next: Version one model artifact and record how you would roll it back.