Projects¶
Every few weeks the lectures stop and you build something. Projects are where the ideas stop being words on a page and start being decisions you have to defend.
They work the same way each time. You get a spec, you work in a small group, you produce something that runs, and then you stand in front of the cohort and answer questions about it. The questions are published with the spec, so you know from day one what you will be asked.
-
Project 1: Rock, paper, scissors
Build an image classifier with no code, break it on purpose, and explain what it actually learned.
Two sessions Groups of 3
-
Project 2: One page, one finding
Answer a real question about a real dataset, then fit the entire answer onto a single page.
Three weeks Groups of 3 or 4 19 October
-
Project 3: How sure are you?
Collect your own data from two sources, join it into one table, and answer a question with a range rather than a number.
One week Groups of 3 or 4 16 November
More projects appear here as each block of modules finishes.
How projects are judged
Not by accuracy. A group with a model that scores 60 percent and can explain exactly why will score higher than a group at 100 percent with nothing to say. The number is evidence. The explanation is the work.
The same applies to analysis. A clear finding you can defend beats a beautiful chart of something nobody asked about.