Lecture notes‎ > ‎

Lecture 11

For today you should have:

  1. Read Chapter 8
  2. Prepared for a quiz.


  1. Quiz.
  2. Project report "suggestions"
  3. Bayesian Estimation exercise.
  4. Coin problem.

For next time:

  1. Homework 8.
  2. Read Chapter 9.
  3. Read this post about multiple regression.

Bayesian estimation

We will start with an in-class exercise where we estimate the probability of heads, p, based on evidence.
The prior is a uniform distribution, or actually a discrete approximation of a uniform distribution.
Each person in the room gets a difference value of p and a degree of confidence.
If we get heads, each person updates with
confidence *= p
and if we get tails, the update is
confidence *= 1-p
After each flip we can look at the location and shape of the distribution.
With a large number of flips, the distribution converges on the actual value of p.
Some lessons:
1) It's ok to start with an unnormalized prior, but we do need the hypotheses to be ME and CE (mutually exclusive and collectively exhaustive).
2) You can normalize the posterior after each flip, or leave it until the end.  Same answer either way.
3) The results depend on the prior, so in that sense it is subjective.  But we can often use context to make justified decisions about the prior.
4) With enough data, people with different priors converge, unless the priors are "immune to data."
Subpages (1): Locomotive solutions