Release v0.2.0 · JuliaRCM/ReducedComplexityModeling.jl

ReducedComplexityModeling v0.2.0

The sampler interface changes: sample(sampler, parameters::NamedTuple) is now the primitive and
infers a concrete Table. sample(sampler, p1, p2, …) still works but does not infer. A sampler
now implements _columns(sampler, parameters::Tuple) instead of
sample(::MySampler, ::Vararg{Parameter}). The keys of the NamedTuple must equal the parameter
names, else an AssertionError is raised.
AutoEncoderModel, an exported stub type with no implemented behaviour, has been removed.
TrainingData is now a struct rather than an abstract type, and it is exported.
GeometricIntegratorData and GeometricIntegratorEnsembleData keep their fields and exports but
no longer subtype it. Nothing dispatched on the old abstract type, so no call site changes.

Two new parameter samplers: RandomParameterSampler(n, rng) and QuasiRandomParameterSampler(n)
(a Halton sequence). Both subtype ParameterSampler, ignore stored samples, and work with
sample and the ParameterSpace(sampler, params...) constructor.
HDF5 groups written by save_parameters, h5save(::Parameter) and h5save(::ParameterSpace) now
track link creation order, so keys are preserved as written. The HDF5 compat bound rises to
0.16.11, 0.17.
Training data and training problems are first class.
TrainingData{SpaceType, DataType, SystemType}(time, data[, problem]) tags data with three
orthogonal axes, TrainingProblem(data, residual, parameters) holds a fit, and
train(problem, method::Optim.FirstOrderOptimizer) fits the parameters. The exported names are
ObservableSpace, IntrinsicSpace, TrainingData, TrainingProblem, train and train!.
This adds GeometricBase to the dependencies, with the compat bound 0.14.6, 0.15.

ParameterSpace and NamedTuple(::Parameter...) accept parameters of different element types.
show(io, ::Parameter) now dispatches through Base.show; read_parameters(fpath) passes its
group path positionally; save_parameters(fpath, params) opens a missing file with "cw"; and
h5save(h5, ::Parameter) writes no samples dataset for a Parameter without samples, so the
round trip works.
The DataLoader constructors read the autoencoder keyword as a Bool, so a value such as
nothing or :yes now raises an error rather than silently returning nothing.
The export read_sampling_parameters, which named no definition, has been removed.

Pkg.test() no longer runs the doctests; test/quality/doctests.jl is the doctests group,
which Pkg.test(test_args = ["doctests"]) runs. CI's Doctests job stays their runner.

Merged pull requests:

Import DataLoader from GeometricMachineLearning. (#31) (@benedict-96)
Normalise the data loader to NFC (#33) (@michakraus)
Strip the stubs, the fork and the marker structs (#34) (@michakraus)
Update the Batch doctest output for Julia 1.13 (#35) (@michakraus)
Fix the parameter layer; add random and quasi-random samplers (#36) (@michakraus)
Mixed-type ParameterSpace, inferred sample, HDF5 key order (#37) (@michakraus)
Make the Batch doctests independent of the random stream (part M17a) (#39) (@michakraus)
Migrate the test suite to the test convention (part M17) (#40) (@michakraus)
Remove the empty Open Issues heading from the CHANGELOG (#41) (@michakraus)
Update Documenter requirement to 1.19.0 in /docs (#43) (@dependabot[bot])
Update DocumenterCitations requirement to 1.5.0 in /docs (#44) (@dependabot[bot])
Update DocumenterInterLinks requirement to 1.1.0 in /docs (#45) (@dependabot[bot])
Update Documenter requirement to 1.19.0 in /test (#48) (@dependabot[bot])
Read the DataLoader autoencoder flag as a Bool (#49) (@michakraus)
test: JET report_opt lines for the kernel launchers and kernel bodies (part J3) (#50) (@michakraus)
CI: coverage from Julia 1 on ubuntu, save the cache only from a job that succeeds (#51) (@michakraus)
Move the doctests into their own test group (#52) (@michakraus)
Training data, training problems and train (#53) (@michakraus)

Closed issues: