The ML Development Lifecycle Overview
What you'll learn: How machine learning projects follow an iterative cycle from identifying a problem all the way through deployment and ongoing monitoring.
The Cycle, Not a Straight Line
Building an ML system isn't like following a recipe once and being done. It's more like tending a garden—you plant, observe, adjust, and continuously improve. The ML Development Lifecycle describes this repeating journey through several connected stages.
The Core Stages
1. Problem Definition
Clearly articulate what you're trying to predict or discover. Is this even an ML problem? (Recall lesson 116: when to use ML vs traditional programming.)
2. Data Collection and Preparation
Gather relevant data, clean it, and explore patterns. Remember: ML is fundamentally data-centric (lesson 121).
3. Feature Engineering
Transform raw data into representations your model can learn from effectively (lesson 117: the role of features).
4. Model Selection and Training
Choose an appropriate learning paradigm (supervised, unsupervised, etc.) and train candidate models using optimization techniques like gradient descent.
5. Evaluation
Assess whether your model generalizes well (lesson 118). Does it perform on unseen data, or just memorize training examples?
6. Deployment
Integrate your model into a real system where it makes predictions on new data.
7. Monitoring and Maintenance
Track performance over time. Data distributions shift, patterns change—your model needs updates.
Why It's Iterative
You rarely get it right the first time. Poor results in evaluation send you back to collect more data, engineer better features, or try different models. Even after deployment, monitoring reveals issues requiring another cycle.
Key Takeaway: The ML development lifecycle is an iterative loop through problem definition, data work, modeling, evaluation, deployment, and monitoring—not a one-time linear process.