- The function
plot_continuation_curves
has been changed and now plots usingscatterlines!
(which also makes it much more efficient). To restore the old behavior of only scatter useslines_kwargs = (linewidth = 0,)
as a keyword. - Internal function
ics_from_grid
is now exported.
- Initial conditions given to
basins_fractions
and similar functions have been generalized toAbstractVector
fromStateSpaceSet
. In practice this means that it is now possible to pass in as initial conditions a vector of dictionaries, which allows specifying initial conditions via a mapping of symbolic variables if the dynamical system was made via ModelingToolkit.jl.
AttractorsViaProximity
has been significantly improved: it now allows for a keyworddistance
. This keyword decides how the distance between the trajectory end-point and the attractors is decided. The function has further been simplified and re-uses the existingset_distance
function. The defaultdistance
keeps the previous behavior unaltered.
- Global continuation can now be performed across any arbitrary curve in parameter space. This is something completely novel and we'll likely work on a paper on this!
This is a big release, with (hopefully) nothing breaking, but lots of deprecations!
- New central Tutorial for Attractors.jl. It also highlights how to enrich a continuation output with other measures of nonlocal stability.
- New global continuation algorithm that generalizes RAFM:
AttractorSeedContinueMatch
. - There is now an extendable API for "matchers", ways to match state spaces sets across a continuation. See
IDMatcher
for the API. - New matcher
MatchBySSSetDistance
that does matching used to do in previous versions. - New matcher
MatchByBasinOverlap
that does whatmatch_basins!
used to do. - New matcher
MatchByBasinEnclosure
that is truly brand new. - New plotting function
plot_continuation_curves!
to add additional information to theplot_basins_attractors_curves
type plots. - New exported function
reset_mapper!
to clear all stored information in anAttractorMapper
. AttractorsViaFeaturizing
now always stores the attractors and implementsextract_attractors
.- New plotting function
plot_attractors
.
- Function
continuation
has been deprecated forglobal_continuation
in preparation for a future where both local/linear/tradiational continuation as well as our "attractors and basins continuation" are both provided by DynamicalSystems.jl. match_continuation!
has been deprecated formatch_sequentially!
.- Option
par_weight
is deprecated inFeaturizeGroupAcrossParameter
. Part of the developer team (@Datseris
,@KalelR
) discussed this an concluded thatpar_weight
doesn't make much scientific sense to include. Since it obfuscates the code and the documentation, it is no longer documented but still exported. It will be unavailable in the next breaking release.
- New function
convergence_and_basins_fractions
- improved the algorithm
counting the convergence iterations for
AttractorsViaRecurrences
to give more accurate results, by taking into account the user-provided convergence criteria, and multiplying withΔt
to obtain the time in correct units. - New function
test_wada_merge
to test the Wada property in 2D basins of attraction. - New function
haussdorff_distance
to compute distance between two binary matrices.
The function plot_basins_attractors_curves
can now take a vector of functions for mapping attractors to real numbers. Each makes a new panel with the chosen projection.
Improvements in the minimal_fatal_shock
algorithm:
- Can now chose target attractors to shock towards, enabling computation of the excitability threshold
- Can give in custom metric function for estimating the norm
- Exported alias
excitability_threshold
forminimal_fatal_shock
- New argument
bbkwargs
can be given toMFSBlackBoxOptim
to propagate more keywords to BlackBoxOptimization.jl
- New function
edgetrack
for finding saddles. - New function
convergence_and_basins_of_attraction
for obtaining the iterates each initial condition took to reach the attractor. - New plotting function
shaded_basins_heatmap
.
- The algorithm of
AttractorsViaRecurrences
has been simplified a bit. The action of the keywordmx_chk_loc_att
has been changed which may lead to different results in some usage cases. Nowmx_chk_loc_att
counts how many steps to take after collecting enough recurrences, and this step count is only increasing. - Additional benefit of this change is that incorrect algorithm behaviour can be caught eagerly. Now an error is thrown when we know in advance the algorithm will fail. (This also affects
continuation
withRAFM
) - The documentation of
AttractorsViaRecurrences
has been improved and clarified. Additionally a video illustrating algorithm behaviour has been added. - A renaming of some of the options (keyword arguments) of
AttractorsViaRecurrences
has been done in line with the clarity increase of the algorithm. The following renames are in place and currently deprecated:mx_chk_fnd_att -> consecutive_recurrences
mx_chk_loc_att -> attractor_locate_steps
mx_chk_att -> consecutive_attractor_steps
mx_chk_hit_bas -> consecutive_basin_steps
mx_chk_lost -> consecutive_lost_steps
mx_chk_safety -> maximum_iterations
- New algorithm
GroupViaPairwiseComparison
to group features inAttractorsViaFeaturizing
. Simpler, typically faster and using less memory than DBSCAN, it can be useful in well-behaved systems.
- New algorithm
subdivision_based_grid
. Allows user to automatically construct the grid which simulates subdivision into regions with different discretization levels in accordance with state space flow speed.
- Added support of irregular grids to
AttractorsViaRecurrences
, now user can provide ranges without fixed step along the same axis.
- Matching attractors during the continuation with
RAFM
has been improved and is done by the functionmatch_continuation!
which has two options regarding how to handle attractors of previous parameters that have vanished.
- New algorithm
minimal_fatal_shock
that finds the minimal perturbation for arbitrary initial conditionu0
which will kick the system into different from the current basin.
- Fixed issue where poincare map was not working with basins of attraction as intended (sampling points directly on the hyperplane)
- Some functions have been renamed for higher level of clarity (deprecations have been put in place):
match_attractor_ids!
->match_statespacesets!
GroupAcrossParameter
->FeaturizeGroupAcrossParameter
.
- Attractors.jl moved to Julia 1.9+
- Plotting utility functions are now part of the API using package extensions. They are exported, documented, and used in the examples.
- Our pre-print regarding the global stability analysis framework offered by Attractors.jl is now online: https://arxiv.org/abs/2304.12786
- Package now has a CITATION.bib file.
- New function
rematch!
that can be used aftercontinuation
is called withAttractorsViaRecurrences
, if the original matching performed was not ideal.
- More plotting functions have been added and plotting has been exposed as API. After Julia 1.9 it will also be documented and formally included in the built docs.
- Add option
force_non_adaptive
toAttractorsViaRecurrences
. This option is of instrumental importance when the grid is too fine or when limit cycle attractors are involved.
basins_fractions
no longer returns the attractors. Instead, a functionextract_attractors(mapper::AttractorMapper)
is provided that gives the attractors. This makes an overall more convenient experience that doesn't depend on the type of the initial conditions used.
- Added the
continuation
function and the two typesRecurrencesFindAndMatch
andGroupAcrossParameterContinuation
This is the first release. It continues from ChaosTools.jl v2.9, and hence, the comparison of attractor-related features is w.r.t to that version.
- New attractor mapping algorithm
AttractorsViaRecurrencesSparse
that uses sparse arrays to find attractors of arbitrarily high dimensional dynamical systems, eliminating the main drawback ofAttractorsViaRecurrences
. - Clustering (used in
AttractorsViaFeaturizing
) has been completely overhauled. Now, aClusteringConfig
instances must be created and then passed on toAttractorsViaFeaturizing
. AttractorsViaFeaturizing
no longer has keywords about clustering.- A new function
cluster_features
is exported to the public API. - Multithreading is now an option in
AttractorsViaFeaturizing
. It is enabled by default. - Added a new clause in automatic
ε
estimation inAttractorsViaProximity
for when there is only a single attractor passed in by the user.
- New function
continuation
that calculates basins fractions and how these change versus a parameter (given a continuation method) - New basins fraction continuation method
RecurrencesFindAndMatch
that utilizes a brand new algorithm to continuate basins fractions of arbitrary systems. match_attractor_ids!
has been fully overhauled to be more flexible, allow more ways to match, and also allow arbitrary user-defined ways to match.- New function
match_basins_ids!
for matching the output of basins_of_attraction`. - New exported functions
swap_dict_keys!, unique_keys, replacement_map
used in code that matches attractors and could be useful to front-end users.