Bayes' Rule¶
Formula¶
\[
P(A\mid B) = \frac{P(B\mid A)\,P(A)}{P(B)}
\]
Parameters¶
- \(A,B\): events with \(P(B)>0\)
- \(P(A)\): prior probability
- \(P(B\mid A)\): likelihood
- \(P(A\mid B)\): posterior probability
What it means¶
Updates belief about \(A\) after observing \(B\).
What it's used for¶
- Updating beliefs with evidence.
- Interpreting diagnostic tests.
Key properties¶
- Follows from the product rule: \(P(A,B)=P(A\mid B)P(B)=P(B\mid A)P(A)\)
- Normalization uses \(P(B)\), often computed via total probability
Common gotchas¶
- Confusing \(P(A\mid B)\) with \(P(B\mid A)\) (base-rate fallacy).
- Forgetting to ensure \(P(B)>0\).
Example¶
If \(P(D)=0.01\), \(P(+\mid D)=0.99\), \(P(+\mid eg D)=0.05\), then \(P(D\mid +)=0.0099/(0.0099+0.0495)\approx0.167\).
How to Compute (Pseudocode)¶
Input: event probabilities / joint distribution entries
Output: requested probability quantity
identify the relevant events/variables and required joint/marginal terms
apply the probability identity in the card formula
check denominator/normalization terms are valid (nonzero when required)
return the computed probability
Complexity¶
- Time: Typically \(O(1)\) for a single event computation once required probabilities are available; larger table-based calculations scale with table size
- Space: \(O(1)\) extra space for a single computation
- Assumptions: Probability terms (joint/marginals/conditionals) are already known or computed separately