LESSWRONG
LW

Bayes' TheoremRationality
Frontpage

15

Interactive Bayes Theorem Visualization

by Allen Kim
3rd Mar 2018
1 min read
5

15

This is a linkpost for https://allenkim67.github.io/bayes-demo/
Bayes' TheoremRationality
Frontpage

15

Interactive Bayes Theorem Visualization
3Ben Pace
2Trevor Hill-Hand
2Trevor Hill-Hand
2Allen Kim
1Romex91
New Comment
Rendering 4/5 comments, sorted by
top scoring
(show more)
Click to highlight new comments since: Today at 1:29 PM
[-]Ben Pace7y30

Promoted to frontpage.

Reply
[-]Trevor Hill-Hand7y20

I love this! I can imagine myself using this as an ad hoc calculator, especially when I need to explain my estimates to someone unfamiliar with probability. For that purpose it might be useful if I could fill in my own labels for all the variables. So instead of "E1/E2/..." the page could literally display "Drew a black marble/drew a white marble/..." or whatever I put in.

Reply
[-]Trevor Hill-Hand7y20

Snapping to whole percentages might be better too. Having a displayed value of "87.23%" overstates how much control I have in choosing a specific value, given how the sliders work and how finicky per-pixel selection is.

Reply
[-]Allen Kim7y20

Hey thanks for the feedback! After playing around with it more myself I agree with you about the percentages. I think ideally for more precise values it would have a manual input option, but rounding to whole values for now would be better. I also like the label changing suggestion.

Reply
[-]Romex913y10

A simplified version of the page for interpretation of medical test results:

https://romex91.github.io/sick-or-healthy/

Reply
Moderation Log
Curated and popular this week
4Comments

This visualization was inspired by the Julia Galef video that I linked to in the opening paragraph. One thing I liked about that video is that it represents probability space as a box which, for understanding Bayes' rule, I think is much more intuitive than the circular venn-diagram approaches. Some improvements that I think this visualization offers is showing the different probabilities updating in relation to each other, generalizing to more than two events for both priors and likelihoods, and explicitly showing the discarding of conditional probabilities of unobserved events and normalization of the posteriors.