move to SSSets v1 and DynamicalSystemsBase v3

[Datseris]

Move to StateSpaceSets.jl v1. Only mass renames, nothing else changed.

[kahaaga]

The tests fail because StateSpaceSet doesn't have an implementation for Base.IteratorSize.

[Datseris]

(in a meeting, if you can do PR otherwise i do later)

[kahaaga]

I don't have write access to StateSpaceSets, so it'll probably be quicker if you do it later

[kahaaga]

The latest version of Combinatorics, both indicated by the github repo and on JuliaHub is v1.0.2, so indicating 1.0.3 won't work

[Datseris]

I am so confused. How do we use datasets in this package that I have always errors that the StateSpaceSets.jl test suite doesn't get? I don't get it. What is this special syntax that we use...?

[kahaaga]

I am so confused. How do we use datasets in this package that I have always errors that the StateSpaceSets.jl test suite doesn't get? I don't get it. What is this special syntax that we use...?

You didn't change anything but the name when renaming?

I can check out the branch and see if I can figure out what's going on, but won't have time to do it before tomorrow

[Datseris]

I've finally found and fixed the problem. For fucks sake that was so sloppy (from my part)

[Datseris]

(locally, tests pass except the transfer operator tests that also fail on main. will push when DynamicalSystemsBase 3.0 is also tagged, as I updated that here as well)

[Datseris]

I just submitted DynamicalSystemsBase v3 for tag, so I need to move towards the release of DynamicalSystems.jl v3.0. I will try to solve the TO tests here but if they do not get solved, I will probably add an error message to the creation to TO saying that "Currently doesn't work" and proceed with tagging.

(I need this package to have v1 of StateSpaceSets.jl)

[Datseris]

Unfortunately I am not familiar enough with the internals of TO to solve this. The question is: were the tests of TO ever passing when moving to v2.0?

[Datseris]

The problem is here:

# The L points visits a total of L bins, which are the following bins (referenced
    # here as cartesian coordinates, not absolute bins):
    visited_bins = map(pᵢ -> encode(encoding, pᵢ), pts)

visited_bins contains -1 as entry.

[kahaaga]

Unfortunately I am not familiar enough with the internals of TO to solve this. The problem is here:

I also can't do this from the top of my head. I need to sit down and understand why the -1 occurs.

The question is: were the tests of TO ever passing when moving to v2.0?

They were working for what was supposed to be 2.0 at some point, but stopping working after the binning refactoring.

For now, I'm okay with just adding a "not working" at the moment. We can test this explicitly by a @test_throws WhateverError probabilities(TransferOperator(), x).

Because of this error, the TransferOperator estimator isn't implemented for CausalityTools yet either, so should be okay to just error it for now.

[Datseris]

I solved the problem!!!

It is because we were not using the precise variant of the encodings.

[Datseris]

Should we make the precise = true the default? It is 4x slowdown though.

[Datseris]

I've changed the constructor of RectangularBinning(::Int) to use 2 times next float. THat was the only binning that was erroring. Now we can leave things as is. I'll merge and tag now.

(after tests pass)

[kahaaga]

Should we make the precise = true the default? It is 4x slowdown though.

Yes, do that! And perhaps open an issue, so we can remember to optimize this too.

Could you just do a quick test that you get approximately the same entropy for the ValueHistogram and TransferOperator with the same binning, so we're sure nothing is broken?

b = FixedRectangularBinning(0:0.2:1.0)
x = Dataset(rand(100000, 3))
abs(entropy(TransferOperator(b), x) - entropy(ValueHistogram(b), x) ) < 0.1 # or something like that

[Datseris]

okay I am also updating eveyrthing to DYnamicalSystemsBase 3.0 in thsi PR

[Datseris]

oh god fingers crossed!

[Datseris]

I've tested everything locally, and built docs locally, all is ok.

[kahaaga]

I've tested everything locally, and built docs locally, all is ok.

Perfect. If the tests and doc build are successfull, then go ahead and merge!

[codecov[bot]]

Codecov Report

Merging #249 (f2eb069) into main (e2cf0c8) will increase coverage by 0.03%. The diff coverage is 86.20%.

@@            Coverage Diff             @@
##             main     #249      +/-   ##
==========================================
+ Coverage   81.03%   81.06%   +0.03%     
==========================================
  Files          49       49              
  Lines        1234     1236       +2     
==========================================
+ Hits         1000     1002       +2     
  Misses        234      234              

Impacted Files	Coverage Δ
src/encoding_implementations/ordinal_pattern.jl	86.79% <ø> (ø)
src/multiscale.jl	0.00% <0.00%> (ø)
src/probabilities.jl	89.28% <ø> (ø)
src/probabilities_estimators/value_histogram.jl	87.50% <ø> (ø)
..._estimators/transfer_operator/transfer_operator.jl	75.20% <66.66%> (ø)
...rc/encoding_implementations/rectangular_binning.jl	90.24% <100.00%> (+0.24%)	:arrow_up:
src/entropies_estimators/nearest_neighbors/Gao.jl	100.00% <100.00%> (ø)
...rc/entropies_estimators/nearest_neighbors/Goria.jl	100.00% <100.00%> (ø)
...estimators/nearest_neighbors/KozachenkoLeonenko.jl	100.00% <100.00%> (ø)
.../entropies_estimators/nearest_neighbors/Kraskov.jl	100.00% <100.00%> (ø)
... and 7 more

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