Hand in your capstone project.

What to hand in

1. model.pt — the final model (~45 MB). If it is too big to send, say how to recreate it: the exact capstone.py command and the seed.
2. report.md — filled in, and read against the rubric of step 6.
3. The experiment .json files — each run's summary, and the output of choose.py.
4. A demo: five photos that are in neither data/ nor test/ (e.g. just taken), run through predict.py. Show the output and mark which ones the model got right and which not.

Check

• python -m unittest passes for every part of the project
• the choose.py output has at least two runs
• the report says how many times test/ was looked at
• the published number is from new images, not from validation
• the report has at least one place where something did not help or remains unknown

Two questions at the end

Write one paragraph each:

1. The first lesson's promise. In the first lesson we promised you would watch one number go from "guessing" to "works". Compare with your real story: where did it jump, where did it stall, where did some idea not help? Which of those steps did you judge wrongly in advance?
2. What you would do differently if you had one more week — not "more data", but concretely: which one experiment, and what should it show?

Tip. If your results are worse than you hoped, do not rush to "fix" the report. A good report with 80 % and clearly described mistakes is worth more than a pretty number you cannot defend.