v0.2.0

Sindbad v0.2.0

Diff since v0.1.1

Breaking changes

• Breaking change: major refactor of the machine-learning and hybrid workflow APIs, including new pullback/update structure and reorganization of ML-related functions and extensions. Existing hybrid ML experiment scripts and code will need to be updated to the new API.
• Breaking change: removal of the large exp_fluxnet_hybrid and legacy z-legacy_exp_hybridNN example suites (and their associated settings/scripts), which may break users relying on those exact example paths or files.

Other changes

• Bump Sindbad to v0.2.0 with updated compat bounds for OmniTools and SindbadTEM.
• Clean up and simplify the examples directory, focusing on core workflows.
• Introduce and extend AD/ML backends via Sindbad* extension modules (Flux, Lux, Optimisers, ForwardDiff, Zygote, PolyesterForwardDiff, PreallocationTools).
• Improve configuration, output handling, and path logic (e.g. avoiding abspath on http/s3 paths and fixing output base paths).
• Expand and update tests for machine-learning gradients and AD methods to cover the new design.

Merged pull requests:

• compat and url updates (#58) (@lazarusA)
• update AD ext methods (#60) (@lazarusA)
• add print_info to mlModels (#61) (@dr-ko)
• Sk/v0.1.2 (#62) (@dr-ko)
• Farquhar models (#63) (@dr-ko)
• updated and added methods and processes for APAR and Conductance (#64) (@relghawi)
• Don't apply abspath on http and s3 path strings (#67) (@felixcremer)
• refactor pullback and update methods (#68) (@dr-ko)
• fix output path issues based on root env (#69) (@dr-ko)
• refactor covariates and hybrid experiments (#70) (@dr-ko)
• v0.1.1 to v0.2.0 (#71) (@lazarusA)

Closed issues:

• Stable docs link is broken (#48)
• @unpack_nt should unpack all slots of NamedTuple if there not assignments. (#49)
• move hybrid experiment (#59)
• outpath setting is ignored by forward run (#66)