BayesianBlocks

Bayesian Blocks algorithm described in Scargle 2021.

partition(data, logfitness=:cash, logprior=:p0, progress=false, ...)

Return an array of optimal change points for a set of one-dimensional data (observations or an histogram of them). This is the implementation of the bayesian blocks algorithm, as outlined in ^[1].

Arguments

• data: numeric array or a StatsBase.Histogram
• logfitness: log of the block fitness function to be used, choose between [:cash]
• logprior: log of the prior distribution on the number of blocks to be used, choose between [:gamma, :p0]
• gamma, p0...: set the parameter value for the specified prior distribution
• progress: display a progress bar for long computations

Example

using Distributions, StatsBase, Plots, LinearAlgebra

data = vcat(rand(Normal(0),1000),rand(Cauchy(5),1000))
data = data[(data .> -5) .& (data .< 10)]

h = fit(Histogram, data, -5:0.1:10)

# choose to use all data or an histogram of it!
hb = fit(Histogram, data, BayesianBlocks.partition(data))
hb = fit(Histogram, data, BayesianBlocks.partition(h))

plot(data, normalized=true, st=:stephist, nbins=1000)
plot!(normalize(hb), st=:step, w = 3)

Performance tips

You can convert your data container to a less precise representation to improve the performance a bit, e.g.

x::Array{Float32} = [1.1, π, (√5-1)/2]