- Error when checking the number of streamlines in the weighted LASSO function
- Error in verifying voxel correspondence between the lesion mask passed through the dictionary and the ISO weights map
- Error when checking the number of streamlines in the grouping structure for group LASSO regularization (Fixes #154)
- Added check on empty groups when setting group LASSO regularization (Fixes #153)
- Tests fail using numpy >= 2.0.0
- Typo in test script and reference results
- Names of the output files of the "lesion" model
- Error when calling "fit" again after "save_results"
- Minor fixes and improvements
- Lesion mask support using ScalarMap models.
- Debias option in
fit()to remove bias from the estimated coefficients when using regularization (e.g.COMMIT2). - Public
dictionary_infoattribute onEvaluationfor reuse (e.g. lesion masks, blur settings, thread counts). - Two options to change the
argsortalgorithm (sortDictionaryWithandsortThreadsWith)
VolumeFractionsmodel now is calledScalarMapfor clarity.- Refactored
Evaluationto load and storedictionary_infoonce and reference it inload_data(),set_model()andsave_results(). - Bumped package version to 2.4.0 in
pyproject.toml.
- Logger prints
- Revert changes when save results in
results.pickle - Add checks on
group_weights_extra
- Added support for Windows (requires the
pthread-win32library) - Precompiled wheels for Windows, MacOS, and Linux are now available on PyPI
operator.pyxno more compiled at runtime- Restict the
numpyversion to<2.0.0
- Improved output when running from Jupyter notebooks
- Fix
set_threadisotropic compartment voxel distribution
- Add
do_reweightingparameter tosave_results() - Add logging module to handle console output
- Work In Progress (WIP) models
- Different regularizers
- Merged PR #127: [FIX] Operator is not recompiled when some parameters are changed
- Merged PR #129: [FIX] Fista update
- Bump dependencies
- Add
dtype=np.object_to inhomogemeous shape arrays - Fix
egg_baseoption insetup.cfg(create thebuildfolder before runningsetup()) - Fix
trk2didtionarycrash when run from Jupyter notebooks (removed C-level stdout from Jupyter notebook)
- Default
ndirs=500incore.generate_kernels()and intrk2dictionary.run() - Expire the deprecated
ndirsparameter inamico.core.setup() - Expire the deprecated
filename_trkandgen_trkparameters intrk2dictionary.run() - Removed unused parameter
ndirsfromtrk2dictionary_c.cpp() - Build output goes into
build - Switched to proprietary license (see
LICENSEfile)
-
Added clustering in trk2dictionary.run()
- geometry-based clustering using
blur_clust_thrparameter - anatomical information streamlines clustering based on their endpoints using
blur_clust_groupbyparameter
- geometry-based clustering using
-
Added trk2dictionary.run() parallel computation using
n_threadsparameter -
Changed internal streamlines representation using
blur_core_extentandblur_gauss_extentparameters -
Added possibility to keep temporal files created by parallel trk2dictionary.run() and clustering using
keep_tempparameter -
Parallel compilation
- Bugfixes
- Build output goes into 'build'
- Require
dmri-amico>=1.3.2,<2.0.0
- Solve #121
- Invalid command 'bdist_wheel' error when install with pip<=20.0.2
- Ensure non-negative values in the 'y' vector before fit
- 'replace_bad_voxels' parameter in 'load_data()' to replace NaN and Inf values in the signal
- Problem with very anisotropic voxelsizes
- generate_kernels: forcing 'ndirs' to 1 in case of 'VolumeFraction' model
- core.setup: removed 'ndirs' parameter, now precomputes all directions
- Shortcut 'commit.core.setup' to 'commit.setup'
- Improved the blur functionality
- Option 'coeffs_format' to 'save_results()'
- Install information are stored (and taken from) commit/info.py
- Missing dependencies
- Option 'get_normalized' to 'save_coeffs()'
- Possibility to specify a voxel confidence map
- Set license: BSD-3-Clause License
- Indices of groups in case of streamlines that were pre-filtered during trk2dictionary
- Loading of nii data using (np.asanyarray( nii.dataobj )) in core and trk2dictionary
- confidence_map from float64 to float32
- Length of short segments to neglect
- Streamline weights, in case of blur, are properly scaled
- operator.pyxbld: Changed the condition to create a new operator
- trk2dictionary.pyx: Check that the tractogram exists before trying to load it and remove the try section
- trk2dictionary.run(): fixed bug with blur parameters and computing the blur
- core.pyx: Add to the function build_operator the parameter build_dir
- core.pyx: The function build_operator checks if the LUT configuration changed before build a new operator
- verbose variables in core.pyx and solvers.py changed to be boolean
- trk2dictionary.run(): removed 'points_to_skip' option
- Option to set one single direction in the resolution of the LUT
- store model parameters to results.pickle
- trk2dictionary.run(): check for invalid parameters passed to the blur
- operator.pyxbld: Changed the condition to create a new operator
- COMMIT version is stored in results.pickle
- COMMIT version is stored in output NIFTI files
- trk2dictionary.run(): bug in the blurring functionality
- trk2dictionary.run(): 'blur_sigma' defaults to 0
- COMMIT_debugger: compilation problem
- COMMIT_debugger: wrong visualization in Linux
- Moved the documentation to the Wiki
- Updated the installation guide
- trk2dictionary.run(): removed 'gen_trk' option
- save_results(): removed 'save_coeff' and 'save_opt_details' parameters
- save_results(): now saving only streamline_weights.txt (not anymore xic.txt, xec.txt, xiso.txt)
- load_dictionary(): renamed 'use_mask' to 'use_all_voxels_in_mask'
- Removed unused 'dictionary_ndirs.dict' file
- trk2dictionary.run(): 'min_fiber_len' defaults to 0.0 for backward compatibility
- added 'get_coeffs()' function to get all estimated coefficients
- save_results(): added 'stat_coeffs' parameter for saving streamline weights
- trk2dictionary.run(): added 'max_fiber_len' parameter to discard long streamlines
- load_data(): added 'b0_min_signal' to discard voxels with very low signal
- Modify setup.py and fix spams dependencies
- Improvements to the COMMIT_debugger.
- Adapt demos to use d_perps instead of ICVFs for setting model parameters.
- Bug when the selected model has EC compartments but no peaks are provided (in trk2dictionary).
- Parameter 'min_fiber_len' in trk2dictionary to discard streamlines shorter than a given length in mm.
- Bug when 'points_to_skip' was higher then streamline length.
- Few corrections to docstring of trk2dictionary.
- Added colorized output. NB: needs AMICO 1.2.0 or above.
- Added possibility to save the predicted DW-MR signal in save_results.
- Minor cleanup.
- Check if dictionary (upon loading) and data have the same geometry.
- Bug while saving coefficients in save_results.
- Improved the loading of the streamlines in trk2dictionary
This version of COMMIT is not compatible with AMICO v1.0.1 of below. If you update COMMIT to this version, please update AMICO to version 1.1.0 or above.
- Changelog file to keep tracking of the COMMIT versions.
- Added compatibility with low resolution LUTs.
- Nothing.