6. Mini Project: Train a Classifier for a Real Task¶
Put the whole unit together on a project of your own choosing.
Requirements¶
- Pick a real, specific task — not just "two random objects," but something with an actual purpose (sorting recycling types, identifying different plant leaves, telling apart two specific tools).
- At least 3 classes, each with varied, non-identifying uploaded samples (Lesson 1's rule still applies).
- Train, then test with new examples your model has never seen (Lesson 2).
- Diagnose and fix at least one real problem you actually hit — don't just stop at the first working version (Lesson 3).
- Report your model's actual accuracy, tested honestly, not just "it works."
- Note one limitation — a case your model would likely get wrong, and why (Lesson 4).
Try it
Complete all six requirements above and write a short summary: what you built, how well it actually performed, and the one limitation you identified.
Tip
"It works" is not a finished project in this course — "it works, here's how well, and here's what would still trip it up" is. That honesty is exactly the skill Unit 1 was building toward.