flimkit 0.14.0__tar.gz → 0.14.1__tar.gz

{flimkit-0.14.0 → flimkit-0.14.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: flimkit
-Version: 0.14.0
+Version: 0.14.1
 Summary: Fluorescence lifetime imaging analysis: reading, fitting, phasors and stitching
 Author-email: Alex Hunt <alexander.hunt@ed.ac.uk>
 License-Expression: MIT

{flimkit-0.14.0 → flimkit-0.14.1}/flimkit/_version.py

@@ -1,4 +1,4 @@
-__version__ = '0.14.0'
+__version__ = '0.14.1'
 fitter_version = '24'
 app_name = 'FLIMKitDEV' if '+dev' in __version__ else 'FLIMKit'
 
@@ -49,5 +49,6 @@ Need to add Phasor once added to the codebase, and update version to 1.0. And ad
 0.13.7 -> Fit ROI Decay saves its result to the session file as soon as it finishes; the lifetimes it wrote into a region used to be lost if FLIMKit closed before something else saved the session. The guide says what an ROI re-fit uses, and the QuPath extension installs from its catalog. No fitted number changes, so fitter_version stays at 21.
 0.13.8 -> PNG export matches the preview: the lifetime map uses the FLIM min, max, gamma, colormap and τ weighting, the µm scale bar is drawn, and the colour scales can be saved beside the images. The scale bar and the OME-TIFF pixel size had both been missing for every PTU, because the export read ImgHdr_PixRes and scaled it from metres, where PTU files carry ImgHdr_PixResol in µm. The TIFFs a fit writes now carry the pixel size too, scaled for binning. Reopening a saved fit shows its residuals again, rebuilt from the saved fit when the session has no fitted curve. A project folder no longer lists its own project.json as a FLIM LABS file. Code merged after a release builds as FLIMKitDEV, a dated pre-release on GitHub only, and every release carries the documentation as a PDF. A .photons IRF is binned on the data's grid instead of at 256 bins and padded, which made it 4x too narrow against 1024-bin data, so fitter_version goes to 22. Fits with any other IRF are unchanged.
 0.14.0 -> free-tau per-pixel fits run in numba instead of scipy once per pixel, on the CPU and on NVIDIA GPUs through numba.cuda, and on a CUDA or ROCm card through torch as a batched Levenberg-Marquardt, so fitter_version goes to 24. The @fit_backend plugin hook lets an add-on run the per-pixel fits on its own C or CUDA code; flimkit-accelerator-plugin is the worked example and can speed free tau up considerably, though nothing needs it. Fixed two fitting bugs: stitched and batch ROI fits paired each tile's amplitudes with the wrong lifetimes, and CPU free-tau fits used the wrong IRF when the IRF had a fitted tail and the background was fitted. Summed fits report standard errors and, by default, bootstrap errors. Export all, phasor PNG export, an in-app update check and update.py. FLIMKitDEV is built from dev.
+0.14.1 -> the compiled desktop app starts on Linux again: PIL._tkinter_finder is bundled, so the matplotlib toolbar no longer dies on a missing module. The installation check accepts an add-on fit backend that solves every free-tau pixel itself instead of demanding that some reach the scipy solver. Releases now publish to PyPI on their own: a release created by the build workflow does not trigger the publish workflow, so the build dispatches it. No fitted number changes, so fitter_version stays at 24.
 1.0.0 -> release version with all core features implemented and tested
 1.5.0 (or 2.0) -> validation of fitting results with known fluorophores.'''

{flimkit-0.14.0 → flimkit-0.14.1}/flimkit.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: flimkit
-Version: 0.14.0
+Version: 0.14.1
 Summary: Fluorescence lifetime imaging analysis: reading, fitting, phasors and stitching
 Author-email: Alex Hunt <alexander.hunt@ed.ac.uk>
 License-Expression: MIT