Release Notes#

v0.5.4 (2026-01-05)#

🐞 Bug Fixes#

  • modelfit: allow passing parameters with expressions as dictionary (59b20a4)

v0.5.3 (2026-01-01)#

🐞 Bug Fixes#

  • io: fix performance regression in v0.5.2 (d8cf3f9)

⚡️ Performance#

  • io: also patch model module for faster saving (62b9c40)

v0.5.2 (2026-01-01)#

🐞 Bug Fixes#

  • io: make saving fit results reentrant and thread-safe (faeb7d7)

v0.5.1 (2025-11-25)#

⚡️ Performance#

  • io: greatly speed up saving fit results (098200c)

    Speeds up repeated serializations of the same callable by caching.

v0.5.0 (2025-11-23)#

✨ Features#

  • modelfit: make results picklable (#25) (eb8a1f4)

    When using dask to parallelize model fitting, the generated dask graph may contain dynamically defined functions or that are not picklable. This commit refactors the fit function to a module-level function, ensuring that the results can be pickled.

v0.4.1 (2025-11-13)#

🐞 Bug Fixes#

  • modelfit: fix dask detection logic (a5b74d1)

v0.4.0 (2025-11-13)#

✨ Features#

  • modelfit: add support for providing weights as DataArrays (76d13e5)

    Weights can now be passed as DataArrays to modelfit, which are broadcasted to match the data being fitted.

v0.3.0 (2025-09-07)#

💥 Breaking Changes#

  • Using modelfit with progress=True now requires the package to be installed with the progress optional dependency group, like pip install xarray-lmfit[progress]. (d7324c9)

  • While adding dask support, this release drops support for rudimentary joblib-based parallelization across multiple data variables; this removes the parallel and parallel_kw arguments to modelfit. Use dask arrays as an alternative. (d3f90df)

✨ Features#

  • modelfit: Add rsquared to modelfit_stats by @newton-per-sqm (#19) (e5a8a1e)

    Co-authored-by: Pascal Muster Pascal.Muster@infineon.com

  • modelfit: properly support dask and drop support for joblib-based parallelization (d3f90df)

    modelfit now supports dask arrays properly with minimal serialization overhead.

♻️ Code Refactor#

  • modelfit: make tqdm an optional dependency (#20) (d7324c9)

    The tqdm package which provides the progress bar when progress=True is now an optional dependency. If not installed, passing progress=True to modelfit will now result in an error.

v0.2.3 (2025-06-11)#

🐞 Bug Fixes#

  • sort dimension order in output to maintain original order (c2a683f)

  • io: correctly handle saving and loading fit results output from Dataset.xlm.modelfit() (77acadf)

v0.2.2 (2025-04-28)#

♻️ Code Refactor#

  • add quotes to type hints for deferred loading (#10) (e0b515c)

v0.2.1 (2025-04-14)#

⚡️ Performance#

  • delay importing lmfit until needed (#9) (6773d03)

    This improves initial import time.

v0.2.0 (2025-04-08)#

✨ Features#

  • modelfit: allow the user to manually specify parameters to include in the fit result (8e6f1a6)

    This also allows for complex models with many parameters given as expressions.

v0.1.3 (2025-03-10)#

🐞 Bug Fixes#

  • allow lower versions of dependencies (139df09)

v0.1.2 (2025-03-08)#

🐞 Bug Fixes#

  • lower numpy min version to 1.26.0 (a9b4928)

v0.1.1 (2025-02-27)#

🐞 Bug Fixes#

  • avoid modifying the original dataset in save_fit (a3157c0)

♻️ Code Refactor#