AIF-C01 notes
Developing ML solutions

Machine Learning Development Lifecycle

The ML lifecycle is the end-to-end process of developing, deploying, and maintaining models. Its phases, in order:

  1. Business goal identification: stakeholders set the value, budget, and success criteria (KPIs).
  2. ML problem framing: turn the business problem into an ML problem.
  3. Data processing: collection, preprocessing, and feature engineering.
  4. Model development: training, tuning, and evaluation.
  5. Deployment: inference and prediction.
  6. Monitoring
  7. Retraining

It needs collaboration between product managers, developers, data scientists, and engineers.

Worked example: Amazon call center routing

PhaseWhat Amazon did
Business goalReduce call transfers caused by a menu that routed customers to the wrong agent
Problem framingPredict the agent skill a call needs: multiclass classification using supervised learning on historical calls
Data collectionFeatures such as recent orders, Kindle ownership, and Prime membership
Preprocessing and visualizationCleaned the data, merged similar labels (all Kindle skills into one), and explored label distributions
TrainingSplit labeled data into training, validation, and test sets, typically 80/10/10 or 70/15/15. Never evaluate on training data
TuningHyperparameter optimization, such as learning rate (too fast never converges, too slow takes too long), plus more feature engineering
Evaluation and deploymentTransfers dropped and customer experience improved

Feature engineering creates and transforms the input variables the model learns from. The model only learns from what you show it.

On this page