46. Deep Learning vs Classical ML: When to Use Which
46.4 Cost and Data Size
Why. Cost and data size decide what a Pune startup can ship this month.
What. Classical models train in seconds on 10k rows. Deep nets may need more labelled images/text and longer epochs. Cloud GPUs cost money — start local CPU demos here.
How.
Decision prompts
- How many labelled rows/images do we really have?
- What is the acceptable training time?
- Do we need an explainable score for auditors?
- Pick the simplest model that meets the metric (Chapter 37).
print("If labelled images < few hundred, prefer transfer learning later or more data — not a giant net from scratch.")