Skip to content

CHANGELOG.md

Latest commit

 

History

History
354 lines (227 loc) · 13.6 KB

CHANGELOG.md

File metadata and controls

354 lines (227 loc) · 13.6 KB

Changelog

TL;DR — Full version history for Pyneapple. Each entry covers Added, Changed, Fixed, and Removed changes per release. Versions v1.6.0–v1.6.2 are intermediate releases between the v1.5.6 and v1.7.0 tags. Pre-releases (v0.6.0–v0.7.3) are noted as such.

[Unreleased]

Added

  • FitResult public container and internal _PixelFitResult dataclass introduced in src/pyneapple/result.py; all fitters now populate results_ with a FitResult after fitting and FitResult is exported from the top-level pyneapple package

Changed

  • Minimum supported Python version lowered from 3.12 to 3.9
  • tomli>=2.0 added as a conditional dependency (python_version < "3.11") to backport the stdlib tomllib module
  • Ruff target-version updated to py39; Black target-version extended to cover py39–py312
  • IDEALFitter.step_tol type changed from list[float] / np.ndarray to dict[str, float] keyed by model.param_names, consistent with how p0 and bounds are specified; TOML configs update from a flat list to a [Fitting.ideal.step_tol] sub-table
  • numpy, scipy, and nibabel dependencies split into Python-version-specific ranges (<3.10, >=3.10,<3.11, >=3.11) to ensure compatible versions are resolved across all supported Python releases

Fixed

  • import tomllib in src/pyneapple/io/toml.py replaced with a try/except ImportError shim that falls back to tomli on Python < 3.11
  • isinstance(value, h5py.Group | h5py.Dataset) in src/pyneapple/io/hdf5.py replaced with the tuple form isinstance(value, (h5py.Group, h5py.Dataset)) — runtime | union in isinstance() requires Python 3.10+
  • Missing from __future__ import annotations added to src/pyneapple/fitters/ideal.py — without it, X | Y union annotations in function signatures were evaluated at class definition time and raised TypeError on Python 3.9
  • ConstrainedCurveFitSolver: fraction_constraint=True now correctly raises ValueError when fit_reduced=False (full-mode signals are not normalised, so the hard constraint sum(f_i) <= 1 is not physically meaningful) and works as intended with fit_reduced=True or fit_s0=True (normalised / amplitude-scaled reduced models)
  • IDEALFitter: cubic-interpolated parameter maps are now clamped to global solver bounds before step-wise bounds are derived, preventing out-of-range initial guesses caused by interpolation overshoot
  • IDEALFitter: a warning is now logged and all voxels are fitted when no voxels survive the segmentation threshold at a coarse IDEAL step, preventing a silent collapse of the multi-resolution chain
  • IDEALFitter._validate_step_tol() now raises a ValueError whose message contains "step_tol" when the dict keys do not match model.param_names
  • configure_logging() no longer attaches a console sink by default; a sink is only added when explicitly requested, preventing duplicate log output
  • CurveFitSolver: passing n_pools=None in multi-threading configuration no longer raises an error

[v2.0.0-beta.2] — 2026-04-22

Fixed

  • SegmentedFitter was missing from pyneapple.fitters.__init__ — it is now properly exported and available as from pyneapple.fitters import SegmentedFitter

[v2.0.0-beta.1] — 2026-04-01

⚠️ Major breaking changes — This release restructures the entire codebase to follow scikit-learn conventions. See release_draft.md for comprehensive migration guide.

Added

  • scikit-learn Style Architecture: Modular structure with Models, Solvers, and Fitters following sklearn estimator patterns
  • Models: Stateless, pure math forward model functions (forward, jacobian, residual) — MonoExpModel, BiExpModel, TriExpModel, NNLSModel
  • Solvers: Curve fitting backends — CurveFitSolver, NNLSSolver, ConstrainedCurveFitSolver
  • Fitters: High-level fitting orchestrators — PixelWiseFitter, SegmentationWiseFitter, IDEALFitter, SegmentedFitter
  • Registry-based plugin discovery via entry points for models, solvers, and fitters
  • pyneapple.io module: New I/O utilities for NIfTI, b-values, TOML configs, Excel, and HDF5
  • pyneapple.utility module: New utilities for spectrum processing, plotting, and validation
  • New CLI tools: pyneapple pixelwise, pyneapple segmented, pyneapple ideal, pyneapple segmentationwise
  • Optional dependencies support: pip install pyneapple[excel,plotting,export]
  • New TOML config format with [Fitting] section replacing old [General] with Class field
  • SegmentedFitter for two-step model fitting workflow
  • NIfTI export with reconstruct_maps and save_spectrum_to_nifti
  • Excel export with save_params_to_excel and save_spectrum_to_excel
  • HDF5 export with gzip compression

Changed

  • TOML config format redesigned: Removed Class field, restructured under [Fitting] section
  • CLI interface redesign: pyneapple pixelwise/segmented/ideal/segmentationwise replaces pyneapple-fitdata
  • Model naming unified: BiExpFitModel → BiExpModel, MonoExpFitModel → MonoExpModel, etc.
  • Logging API: configure_logging() from pyneapple replaces from pyneapple.utils import logger
  • Moved from poetry to uv for package management

Removed

  • pyneapple/parameters/ module entirely: Parameters, IVIMParams, IVIMSegmentedParams, NNLSParams, NNLSCVParams, IDEALParams, BaseBoundaryDict
  • pyneapple/results/ module entirely: Results, IVIMResults, IVIMSegmentedResults, NNLSResults, ResultDict
  • pyneapple/fitting/ module entirely: FitData and all fitting functions
  • pyneapple/utils/ module entirely: logger, plotting, processing
  • radimgarray submodule: RadImgArray, SegImgArray no longer available
  • pygpufit module: Moved to separate package
  • JSON/TOML parameter files: All files in examples/parameters/ replaced with new format
  • Example scripts: ivim_fitdata_script.py, nnls_fitdata_script.py
  • Old fitting modes: fit_pixel_wise(), fit_segmentation_wise(), fit_ivim_segmented(), fit_ideal()

Fixed

  • Various import path corrections and cleanup

[v1.7.0] — 2026-01-19

Added

  • HDF5 data storage and export (3D/4D support, dict import, sparse library for N-D arrays) (#198, #199)
  • NIfTI helper functions and refactored results saving for NIfTI format
  • Dedicated IO module consolidating JSON, TOML, NIfTI, and HDF5 operations
  • fit_opt parameter to track fitting type (pixelwise / segmentation / segmented / ideal)
  • NNLS peak merging and cutoff handling
  • Comprehensive test suites for HDF5, NIfTI, and IO integration

Changed

  • Switched from scipy.sparse to sparse library for N-D array support
  • GPU fitter model retrieval and constraint handling improved (#197)
  • DataExport documentation expanded

Fixed

  • fit_opt defaulting to empty string
  • GPU fitter constraints and import errors
  • NNLS model params not passed properly
  • Shape calculation issues
  • Missing logger output

[v1.6.2] (intermediate — not a published tag)

Added

  • fit_t1_steam option for T1 STEAM correction (#193)
  • Example TOML files for BiExp IVIM and NNLS fitting
  • Fit example scripts and extended example descriptions
  • T1 fit check during results export

Changed

  • IVIM results refactored: unified S0 and fraction extraction method
  • dimstep now always returns np.ndarray
  • Class descriptions and type hints updated throughout

Fixed

  • Segmented test issues after IVIM results rework
  • Additional small-fraction extraction edge case
  • Parameter class invoke error
  • JSON-to-params_file conversion issues

[v1.6.1] (intermediate — not a published tag)

Added

  • TOML parameter file import and export (#168)
  • TOML test suite and import examples
  • TOML documentation
  • Dependencies: tomli, tomli-w, tomlkit

Changed

  • params.json renamed to params.file to support multiple formats
  • Backward compatible with existing JSON parameter files

[v1.6.0] (intermediate — not a published tag)

Added

  • Custom loguru-based logger (#167)
  • Configurable log levels; file output support
  • Logger tests; log level set to ERROR for pytest runs
  • Linux GPUFit library (Ubuntu 22.04 LTS)
  • Model class restructuring: base fit model classes, explicit mono/bi/tri-exponential models, GPU model constructor (#171)
  • Dynamic dict-based parameter handling replacing static file loading (#184)

Changed

  • Replaced print statements throughout codebase with loguru calls
  • reduced renamed to fit_reduced; fit_s0 and fit_t1 properly propagated
  • Model wrapper pattern replaced with direct model classes
  • Moved parameter file fixtures from fitting/ to parameters/
  • *_comment* attributes renamed to *description*
  • tox.ini moved into pyproject.toml

Fixed

  • LOG_LEVEL variable naming
  • Removed unwanted global variables and file output side effects
  • Missing segmented S0 value
  • kwargs overwriting boolean settings

[v1.5.6] — 2025-03-21

Added

  • Full GPUfit integration for CUDA GPU fitting
  • Explicit mono-, bi-, and tri-exponential models (replaces multi-exponential wrapper)
  • New result classes for structured IVIM and NNLS output
  • Individual pixel-by-pixel boundary support (btype) (#197)
  • pytest-mock and pytest-cov; fitting, boundary, and GPU fitter test suites
  • Python 3.13 support

Changed

  • Model wrapper deprecated in favor of explicit model classes
  • Parameter files restructured; files from v1.3.1 or older are no longer supported
  • Segmented fitting reworked with two sub-parameter classes
  • Boundary system reworked to altered dict structure with order handling

Fixed

  • MonoExp equation order (exp(D1) * S0)
  • fit_s0 addition to params2
  • D1 position in first segmented fit pass
  • NNLS results boolean parsing and variable array length

[v1.5.5] — 2025-03-20

Fixed

  • Removed IVIM parameter reorder array — all models now use the same parameter set

[v1.5.4] — 2025-03-12

Fixed

  • S0 and fraction extraction issues for IVIM fitting

[v1.5.3] — 2025-03-07

Fixed

  • Typos, broken conditionals, and fit_model selection issues

[v1.5.1] — 2024-12-11

Changed

  • IDEAL fitting moved to a separate branch; IDEAL files removed from main until stable (#156)

[v1.5.0] — 2024-12-11

Added

  • Full GPU integration for all supported model types
  • kwargs passing for GPU fitter tuning

Changed

  • Fitting models reworked to share a unified interface across CPU and GPU backends (#158)

[v1.4.0] — 2024-12-05

Added

  • GPU fitting via pygpufit for CUDA-capable GPUs (#150)

[v1.3.1] — 2024-11-09

Added

  • RadImgArray implementation for image array handling (#145)

Changed

  • Reconstruction split into UI and fitting layers (#141)
  • Results classes reworked (#146)

Removed

  • UI components extracted from the fitting library

[v1.1.1] — 2024-10-01

Added

  • Load b-values option via context menu (#131)
  • IVIM fixed parameters support (#136)
  • Python 3.9 support (#138)

Changed

  • Fit results converted to custom dict types (#121)
  • fit_data restructured for clarity (#133)

Fixed

  • Save results bug (#127)

[v1.0.0] — 2024-04-29

First public release.

Added

  • Pixel-wise and segmentation-wise IVIM fitting (mono-, bi-, tri-exponential models)
  • NNLS spectral fitting
  • JSON parameter file support
  • Core test suite
  • Basic UI (subsequently removed in v1.3.1)

Notes

  • IDEAL fitting shipped as work-in-progress

[v0.7.3] — 2024-04-17 (pre-release)

Added

  • New package structure with icons and cleaned directory layout
  • NNLS and NNLS-regularised parameter unification (#88)
  • File dialog memory (last_dir property) (#96)

Fixed

  • reg = 0 fitting edge case (#59)
  • Fit area selection (#64)
  • Zero-padding (#89)
  • IDEAL fitting fixes (#85)
  • Various UI and import cleanup (#67, #97)

[v0.7.0] — 2024-02-15 (pre-release)

Added

  • Initial IDEAL fitting support

[v0.6.1] — 2023-12-14 (pre-release)

Minor patch. No release notes available.

[v0.6.0] — 2023-12-13 (pre-release)

Initial pre-release. No release notes available.