Multiscale analysis

[kahaaga]

A suggestion for a multiscale API.

Fixes #102.

Multiscale methods are simple: they apply some entropy/complexity measure on successively more coarse-grained/downsampled time series, and return the computed measure per scale/downsampling level.

I propose the following function to compute multi-scale measures.

multiscale(e::Entropy, alg::MultiScaleAlgorithm, x::AbstractVector, est::ProbabilitiesEstimator;

This function is similar to entropy, but adds a second argument alg, which is some MultiScaleAlgorithm indicating how downsampling/coarse-graining is done. This generalises all the "multiscale-XYZ" methods found in the literature, where XYZ is some entropy, complexity measure, or combinations of various measures.

EDIT: the following is outdated. Multiscale complexity will be taken care of in Complexity.jl.

In the current state of the PR, however, only multiscale entropies can be estimated (i.e. using a combination of some Entropy and a ProbabilitiesEstimator). Other complexity measures can't just as easily be added, because we have no common interface complexity that is analogous to entropy. @ikottlarz, @Datseris should decide on how this interface should look before proceeding with this PR.

Issue #133 tracks the discussion regarding a potential complexity interface. Alternatively, I'll have to implement a unique multiscale_X function for each complexity measure X ∈ [statistical_complexity, sample_entropy, reverse_dispersion, missing_disp_patterns, ...] (which I don't particularly like the idea of, when a solution like #133 would be much more elegant)

[codecov]

Codecov Report

Merging #132 (aa7c7e4) into main (321853b) will increase coverage by 7.82%.
The diff coverage is 89.19%.

❗ Current head aa7c7e4 differs from pull request most recent head 54c1c3e. Consider uploading reports for the commit 54c1c3e to get more accurate results

@@            Coverage Diff             @@
##             main     #132      +/-   ##
==========================================
+ Coverage   76.03%   83.85%   +7.82%     
==========================================
  Files          30       41      +11     
  Lines         776      960     +184     
==========================================
+ Hits          590      805     +215     
+ Misses        186      155      -31     

Impacted Files	Coverage Δ
src/deprecations.jl	0.00% <0.00%> (ø)
src/encoding/outcomes.jl	0.00% <0.00%> (ø)
src/encoding/utils.jl	100.00% <ø> (ø)
...t_entropies/nearest_neighbors/nearest_neighbors.jl	100.00% <ø> (ø)
...abilities_estimators/timescales/wavelet_overlap.jl	90.00% <ø> (ø)
src/probabilities_estimators/count_occurences.jl	44.44% <28.57%> (-55.56%)	⬇️
src/multiscale.jl	33.33% <33.33%> (ø)
src/encoding/gaussian_cdf.jl	91.66% <50.00%> (ø)
src/entropy.jl	71.87% <66.66%> (+3.12%)	⬆️
src/probabilities.jl	74.07% <75.00%> (-0.93%)	⬇️
... and 31 more

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[Datseris]

Oh damn, we really need to prioritize fixing the docbuild so that PRs become easier to review. I will review this later as I need to checkout locally and build docs locally.

[kahaaga]

Oh damn, we really need to prioritize fixing the docbuild so that PRs become easier to review. I will review this later as I need to checkout locally and build docs locally.

What do you mean? The preview docs are building just fine. They're here: https://juliadynamics.github.io/Entropies.jl/previews/PR132/

[Datseris]

What the heeeeeeeeeelll why aren't they displayed in the CI checks as "deployed"??? There is typically one more entry after the docs CI build. Alright, perfect, thanks. Now I can have an actual review here (not right now, later)

[kahaaga]

What the heeeeeeeeeelll why aren't they displayed in the CI checks as "deployed"??? There is typically one more entry after the docs CI build. Alright, perfect, thanks. Now I can have an actual review here (not right now, later)

I have no idea why they're not displayed. They used to before. I'll see if I can decipher what's going on.

[kahaaga]

@Datseris We already agreed on our Zoom call that the proposed API here is ok. I will make sure that all methods follow the type-first approach (#184), then merge this.

[kahaaga]

Tests that fail are unrelated to this PR. Merging.