WeightVectors

AbstractWeightVector <: AbstractVector{Float64}

An abstract vector capable of storing normal, non-negative floating point numbers on which rand samples an index according to values rather than sampling a value uniformly.

FixedSizeWeightVector <: AbstractWeightVector

An object that conforms to the AbstractWeightVector interface and cannot be resized.

WeightVector <: AbstractWeightVector

An object that conforms to the AbstractWeightVector interface and can be resized.

Reallocate w with the size len, compacting w into that new memory. Any elements if w past len must be set to zero already (that's a general invariant for Weights, though, not just this function).

binomial_sample(rng, trials, px, py)

Flip a coin with probability of px//py of coming up heads trials times and return the number of heads.

Has O(trials) expected runtime with a very low constant factor.

Implementation based on Farach-Colton, M. and Tsai, M.T., 2015. Exact sublinear binomial sampling.

binomial_sample_fair_coin(rng, trials)

Flips trials fair coins and reports the number of heads.

Flips up to 64 coins at a time.

initialize_empty(len::Int)::Memory{UInt64}

Initialize a Memory that, when underlaying a Weights object, represents len zeros.

multinomial_sample(rng, trials, weights)

Draw trials elements from the probability distribution specified by weights (need not sum to 1) and return the number of times each element was drawn.

Runs in O(trials * weights), but can be as fast as O(trials) if the weights are skewed toward big weights at the beginning.