MessagePassingRulesTestUtils

Tools for testing message passing rules on their own, without a graph, for packages that define nodes and rules with MessagePassingRulesBase:

• tables of cases for message rules, marginal rules and average energies, each case also run with inputs of other float types, through rule! for an in-place rule and on a poisoned scratch;
• verification of a message rule, and its log scale, against the node's log-density, integrated numerically;
• derivative checks, ForwardDiff through a rule against finite differences;
• the rule-coverage gate, which fails a suite when a rule has no test.

import Pkg; Pkg.add("MessagePassingRulesTestUtils")

using MessagePassingRulesBase, MessagePassingRulesTestUtils, BayesBase, ExponentialFamily, Test

struct Gaussian end   # out ~ Normal(μ, v)
Gaussian(μ, v) = NormalMeanVariance(μ, v)

@define_factor_node(node = Gaussian, type = Stochastic, interfaces = [:out, :μ, :v])

@define_message_update_rule(
    node = Gaussian, target = :out,
    args = (m[:μ]::NormalMeanVariance, m[:v]::PointMass),
    logscale = 0,
    body = (args) -> NormalMeanVariance(mean(args.m[:μ]), var(args.m[:μ]) + mean(args.m[:v])),
)

@testset "Gaussian: out" begin
    @test_message_update_rule(
        node = Gaussian, target = :out,
        cases = [(m = (μ = NormalMeanVariance(1.0, 2.0), v = PointMass(0.5)),) => NormalMeanVariance(1.0, 2.5)],
    )
    @verify_message_update_rule(node = Gaussian, target = :out, m = (μ = NormalMeanVariance(1.0, 2.0), v = PointMass(0.5)))
end

isempty(check_rule_coverage(@__MODULE__))   # true: every rule has a test

• Documentation: https://reactivebayes.github.io/MessagePassingRulesTestUtils.jl/stable/, with two tutorials and a page on reading a failure. make docs builds it locally, into docs/build.
• Tests: make test runs the suite as CI does; make test test_args="name:coverage" selects items by tag, by name or by file. make help lists the other targets: docs, docs-serve, format, check-format, test-fast.
• Depends on MessagePassingRulesBase, BayesBase, Distributions, ForwardDiff and HCubature, and on no engine. A test dependency: nothing at run time needs it. Julia 1.10 or later. MIT licence.