v2.6.0

EnsembleKalmanProcesses v2.6.0

Diff since v2.5.0

Minor changes

• New default variant of EKS (ALDI),
• EKI now has a pseudoinverse fallback, and adds sqrt(eps)*I to covariances to avoid more singular exceptions
• Adds utility to see UKI ensemble before creating an EKP object
• updated documentation for EKI and EKS, (clarifications and latex bugfixes)
• Scalar covariances are now interpreted as UniformScalings not 1x1 matrices
• == for DataContainers

Merged pull requests:

• Bump actions/checkout from 4 to 5 (#515) (@dependabot[bot])
• Fix links in visualize section (#518) (@ph-kev)
• Add ALDI as default form of EKS (#519) (@odunbar)
• bugfix docs image link (#520) (@odunbar)
• Add fallback for linear solve (#521) (@nefrathenrici)
• Add convenience to construct UKI initial ensemble before creating EKP (#524) (@odunbar)
• Add some more non-breaking improvements before new minor release (#525) (@odunbar)
• Update Project.toml to v2.6.0 (#526) (@odunbar)

Closed issues:

• O3.7.1 Momentum-accelerated EKP (#321)
• EKS update does not match docs "ALDI" description (#375)
• Add GNKI to the toolkit (#404)
• O6.X AMIP-calibration-focused developments (#446)
• scalar interpreted as scalar*I, not as 1x1 matrix (#472)
• Improve plotting functionality (#476)
• Bug throwing error in 1D problems with prior constructor (#494)
• Add == for DataContainers (#510)
• Incorrect Latex equations for inversion docs. (#511)
• Formatting of links incorrect (#516)
• Add linear solve methods to hand singular matrices (#517)
• Utility to help create the UKI initial sigma ensemble (#523)

v2.5.0

EnsembleKalmanProcesses v2.5.0

Diff since v2.4.2

Merged pull requests:

• Add compact entry for Makie (#493) (@ph-kev)
• Docs for some of the newest features etc (#496) (@odunbar)
• Add Metadata to Observation object (#498) (@odunbar)
• Bugfix string metadata being made into Chars (#499) (@odunbar)
• typo from docs update (#502) (@odunbar)
• tsvd_cov_from_samples - type stability (#505) (@odunbar)
• Update Makie in compat to v0.23 (#512) (@ph-kev)
• Update Project.toml to v2.5.0 (#514) (@odunbar)

Closed issues:

• Documentation for Impose-prior versions of EKP. (a.k.a infinite-time EKI) (#450)
• Add MakieRecipes to complement PlotRecipes (#475)
• Documentation for SVD utilities and failure-handling improvements (#500)
• Unexpected type from EKP.tsvd_cov_from_samples (#504)

v2.4.2

EnsembleKalmanProcesses v2.4.2

Diff since v2.4.1

Changes

• new error metrics, retrievable with get_error_metrics(ekp) and get_error(ekp) computed from compute_error! (see docstrings for more information)
• new quick-plot tools with Makie (and docs) see Visualizations
• new getters for get_obs(ekp, iteration) and get_obs_noise_cov(ekp,iteration) to help with getting stuff from minibatched observation series

Merged pull requests:

• Add Makie extension for plotting (#482) (@ph-kev)
• compute_error! bugfix, and add some more error metrics (#484) (@odunbar)
• Update Project.toml to v2.4.2 (#490) (@odunbar)

Closed issues:

• with NaNs, the error is computed is NaN (#483)
• Improve getters for learning rate and observation (#487)
• TransformUnscented throws error for large ensemble size (#488)

v2.4.1

EnsembleKalmanProcesses v2.4.1

Diff since v2.4.0

Merged pull requests:

• Update GaussianRandomFields compat, remove Plots dependency (and v2.4.1) (#486) (@nefrathenrici)

Closed issues:

• Unscented(prior) errors when a one dimensional prior is passed in (#485)

v2.4.0

EnsembleKalmanProcesses v2.4.0

Diff since v2.3.1

Merged pull requests:

• New UTKI variant TransformUnscented (#469) (@odunbar)
• SVDplusD, for representing a covariance as a sum of SVD and Diagonal without building it (#473) (@odunbar)
• New compat for Optim >=1.12 (#479) (@odunbar)
• Metadata for ObservationSeries (#480) (@odunbar)
• Update Project.toml to v2.4.0 (#481) (@odunbar)

Closed issues:

• introduce UTKI, high-dimension scaling UKI (#466)
• UKI/UTKI currently do not have update-groups implemented (#467)
• UKI in 1 dimension breaks (#468)
• create SumOfCovariances type that allows obs_noise_cov matrix to be defined as e.g. (SVD + Diagonal + ...) (#474)
• Add hook to add some metedata somewhere (maybe in the observation) (#477)

v2.3.1

EnsembleKalmanProcesses v2.3.1

Diff since v2.3.0

Merged pull requests:

• ETKI (infinite-version) bugfixes and clarity of logic (#470) (@odunbar)
• Update Project.toml V2.3.1 (#471) (@odunbar)

v2.3.0

EnsembleKalmanProcesses v2.3.0

Diff since v2.2.0

Merged pull requests:

• Extend observations handle (T)SVD covariance objects efficiently (#453) (@odunbar)
• New Quantification of member failure, and more flexible NaN handling (#463) (@odunbar)
• Update Project.toml v2.3.0 (#465) (@odunbar)

Closed issues:

• Allow the definition of covariance matrix and inverses via (e.g. low-rank) factorizations, within observation blocks to reduce storage requirements (#452)
• Failure-Handling refinement (a.k.a NaN handling) (#460)

v2.2.0

EnsembleKalmanProcesses v2.2.0

Diff since v2.1.2

Merged pull requests:

• Impose prior option (a.k.a finite-time vs infinite-time EKI) (#444) (@odunbar)
• Improve DMC scheduler scaling with large observation dimension (#462) (@odunbar)
• Update Project.toml to v2.2.0 (#464) (@odunbar)

Closed issues:

• Include Impose-prior versions of EKP. (a.k.a infinite-time EKI) (#437)
• DMC scales poorly with ETKI (#461)

v2.1.2

EnsembleKalmanProcesses v2.1.2

Diff since v2.1.1

Merged pull requests:

• [WIP] bugfix u_idx for minibatch + parameter_groups (#458) (@odunbar)
• Update Project.toml (#459) (@odunbar)

Closed issues:

• Indexing Errors with Parameter Grouping (#457)

v2.1.1

EnsembleKalmanProcesses v2.1.1

Diff since v2.1.0

Merged pull requests:

• Update README.md (#445) (@odunbar)
• Bugfix SinusoidInterface example (#448) (@odunbar)
• Bugfix for bounded above distributions (#455) (@odunbar)
• Update Project.toml V2.1.1 (#456) (@odunbar)

Closed issues:

• Bugs in SinusoidInterface example (#447)
• constrained_gaussian behaves unexpectedly with Inf in bounds (#454)