Machine learning (ML) is a computational discipline for learning patterns from data to make predictions or decisions. In production, an ML system spans problem formulation, experimentation, training, validation, deployment, monitoring, and improvement. section 1.0 section 2.6
Start by deciding whether ML is warranted: simple heuristics may suffice, while complex, high-dimensional data with intricate nonlinear patterns can require a model. Once selected, the problem must be defined with business and technical stakeholders. section 1.1.1
Treat ML development as an iterative workflow:
- Prepare data, train a model, and evaluate its performance. section 1.1.1
- Return to earlier steps when metrics, data quality, or model assumptions expose problems. section 1.1.1
- Orchestrate and automate the workflow to reduce mistakes, accelerate feedback, and support reproducibility across environments. section 1.1.1 section 1.2.1
OCR model
An optical character recognition model can extract identification numbers. The workflow may iterate between data preparation, model training, and evaluation until the system meets its required performance. section 1.1.1
Figure 1.1 The experimentation phase of the ML life cycle
Use the life-cycle diagram to trace the experimentation loop and its repeated transitions between data, training, and evaluation. section 1.1.1
