Re: Aggregating Probability

On Jul 11, 11:20 am, Paul Rubin <ru...@xxxxxxx> wrote:

Neither of the above. You need to obtain data on the frequency with
which E occurs given each possible combination of A, B and C. As an
example, suppose that the probability for E is given by the following table:

Scenario: A only B only C only any other combination
Pr(E|scenario): 0.1 0.9 0.999 any number you want

Your sample data would be consistent with this, assuming that the 10
occurrences of A were A-only events, etc. The conditional probability of
E given all three of A, B and C is any number you want (within the range
[0, 1]).

/Paul

Interesting, thanks. Would either of my answers be correct if we
assumed that A, B, and C were conditionally independent of each other?
I'm trying to understand how to quickly aggregate large numbers of
features, and calculating "each possible combination" would take
forever since it would have O(n^2) performance.
.

Quantcast