LECTURE HUB
Lecture 15 - Priors in practice
🚦 Status🟣 Formatted
🧠 Mastery🐥 Apprentice
📅 Last Review7 Dec 2025
🎯KEY TAKEAWAYS
- A prior can be written as pretend data: I believe this as strongly as if I had already seen 2 successes and 8 failures.
- Updating is then addition. New successes on one pile, new failures on the other.
- A weak prior is overruled by ten observations. A stubborn one barely moves. Choosing how stubborn is the honest part and nobody can do it for you.
- 7 out of 10 on the quiz, on a day when my attention was in a meeting room across the city.
⚡QUICK JOTS
- 18:30 quiz. twenty minutes. two of them I answered the question I had revised, not the one asked
- cannot stop thinking about the committee. they met today
- prior = pretend data. "as if I had already seen 2 passes and 8 fails"
- then just ADD the real ones. 6 and 4 → 8 and 12 → 40 %
- stubborn prior: 20 and 80. same ten trials → 23.6 %. barely moved
- neither engineer is wrong. they disagreed before the data
- write the prior down first, with a date
- problem set 3 due next week. hand it in early for once
☑️HOMEWORK
- Do 📌 Statistics II - Problem set 3 - Priors on paper by hand, all three updates
- Go over the three I dropped in 📌 Statistics II - Bayes quiz
- Add tonight's worked example to 🔭 Bayesian thinking for product bets
- Write down a prior for the scorecard prototype sessions before they start