core/dev.gmitch215.drift.plan/Information

Information

Entropy and expected information gain in micro-bits (1_000_000 is one bit). Every figure is an integer: log2 is exact for powers of two and otherwise within 2 micro-bits of the true value (the error of FixedPoint.ln plus one division rounded to nearest), and each entropy term is rounded to nearest, ties up.

Functions

branches

fun branches(prior: Belief, table: OutcomeTable): List<Branch>

One branch per outcome with its probability and the posterior after seeing it. Bayes over integers: joint = prior times likelihood, then largest-remainder normalization, so both the probabilities and each posterior sum to exactly 1_000_000. An outcome no hypothesis can produce has probability 0 and leaves the prior unchanged.

entropy

fun entropy(probabilities: List<Long>): Long

-sum p log2 p over probabilities (micro-units); zero masses add nothing.

gain

fun gain(prior: Belief, table: OutcomeTable): Long

fun gain(prior: Belief, branches: List<Branch>): Long

Prior entropy minus the expected posterior entropy over branches; never negative.

log2

fun log2(p: Long): Long

log2(p / 1_000_000) in micro-bits for 0 < p <= 1_000_000; zero or negative.