dmri-commit 2.0.0__tar.gz → 2.1.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/LICENSE +1 -1
- {dmri-commit-2.0.0/build/dmri_commit.egg-info → dmri-commit-2.1.0}/PKG-INFO +4 -4
- {dmri-commit-2.0.0 → dmri-commit-2.1.0/build/dmri_commit.egg-info}/PKG-INFO +4 -4
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/build/dmri_commit.egg-info/requires.txt +2 -2
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/core.pyx +267 -50
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/info.py +2 -2
- dmri-commit-2.1.0/commit/models.py +98 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/operator/operator.pyx +1 -0
- dmri-commit-2.1.0/commit/proximals.pyx +198 -0
- dmri-commit-2.1.0/commit/solvers.py +364 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/trk2dictionary/trk2dictionary.pyx +38 -14
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/trk2dictionary/trk2dictionary_c.cpp +87 -10
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/setup.cfg +3 -3
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/setup.py +7 -1
- dmri-commit-2.0.0/commit/models.py +0 -32
- dmri-commit-2.0.0/commit/proximals.pyx +0 -141
- dmri-commit-2.0.0/commit/solvers.py +0 -443
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/MANIFEST.in +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/README.md +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/build/dmri_commit.egg-info/SOURCES.txt +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/build/dmri_commit.egg-info/dependency_links.txt +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/build/dmri_commit.egg-info/top_level.txt +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/__init__.py +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/core.cpp +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/operator/__init__.py +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/operator/config.py +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/operator/operator.pyxbld +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/operator/operator_noLUT.c +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/operator/operator_withLUT.c +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/trk2dictionary/ProgressBar.h +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/trk2dictionary/Vector.h +0 -0
- {dmri-commit-2.0.0 → dmri-commit-2.1.0}/commit/trk2dictionary/trk2dictionary.cpp +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
################################################################################
|
|
2
2
|
COMMIT software license agreement
|
|
3
|
-
Version 2
|
|
3
|
+
Version 2, 24 July 2023
|
|
4
4
|
################################################################################
|
|
5
5
|
|
|
6
6
|
------------------------------------PREAMBLE------------------------------------
|
|
@@ -1,19 +1,19 @@
|
|
|
1
1
|
Metadata-Version: 2.1
|
|
2
2
|
Name: dmri-commit
|
|
3
|
-
Version: 2.
|
|
3
|
+
Version: 2.1.0
|
|
4
4
|
Summary: Convex Optimization Modeling for Microstructure Informed Tractography (COMMIT)
|
|
5
5
|
Home-page: https://github.com/daducci/COMMIT
|
|
6
6
|
Author: Alessandro Daducci
|
|
7
7
|
Author-email: alessandro.daducci@univr.it
|
|
8
|
-
License:
|
|
8
|
+
License: Proprietary license
|
|
9
9
|
License-File: LICENSE
|
|
10
10
|
Requires-Dist: setuptools>=46.1
|
|
11
11
|
Requires-Dist: Cython>=0.29
|
|
12
12
|
Requires-Dist: numpy>=1.12
|
|
13
13
|
Requires-Dist: scipy>=1.0
|
|
14
14
|
Requires-Dist: dipy>=1.0
|
|
15
|
-
Requires-Dist: dmri-dicelib>=1.0.
|
|
16
|
-
Requires-Dist: dmri-amico>=2.0.
|
|
15
|
+
Requires-Dist: dmri-dicelib>=1.0.3
|
|
16
|
+
Requires-Dist: dmri-amico>=2.0.1
|
|
17
17
|
|
|
18
18
|
|
|
19
19
|
========
|
|
@@ -1,19 +1,19 @@
|
|
|
1
1
|
Metadata-Version: 2.1
|
|
2
2
|
Name: dmri-commit
|
|
3
|
-
Version: 2.
|
|
3
|
+
Version: 2.1.0
|
|
4
4
|
Summary: Convex Optimization Modeling for Microstructure Informed Tractography (COMMIT)
|
|
5
5
|
Home-page: https://github.com/daducci/COMMIT
|
|
6
6
|
Author: Alessandro Daducci
|
|
7
7
|
Author-email: alessandro.daducci@univr.it
|
|
8
|
-
License:
|
|
8
|
+
License: Proprietary license
|
|
9
9
|
License-File: LICENSE
|
|
10
10
|
Requires-Dist: setuptools>=46.1
|
|
11
11
|
Requires-Dist: Cython>=0.29
|
|
12
12
|
Requires-Dist: numpy>=1.12
|
|
13
13
|
Requires-Dist: scipy>=1.0
|
|
14
14
|
Requires-Dist: dipy>=1.0
|
|
15
|
-
Requires-Dist: dmri-dicelib>=1.0.
|
|
16
|
-
Requires-Dist: dmri-amico>=2.0.
|
|
15
|
+
Requires-Dist: dmri-dicelib>=1.0.3
|
|
16
|
+
Requires-Dist: dmri-amico>=2.0.1
|
|
17
17
|
|
|
18
18
|
|
|
19
19
|
========
|
|
@@ -1,6 +1,5 @@
|
|
|
1
1
|
#!python
|
|
2
2
|
#cython: language_level=3, boundscheck=False, wraparound=False, nonecheck=False, cdivision=True, initializedcheck=False, binding=False
|
|
3
|
-
from __future__ import print_function
|
|
4
3
|
cimport cython
|
|
5
4
|
import numpy as np
|
|
6
5
|
cimport numpy as np
|
|
@@ -16,6 +15,7 @@ import commit.models
|
|
|
16
15
|
import commit.solvers
|
|
17
16
|
import amico.scheme
|
|
18
17
|
import amico.lut
|
|
18
|
+
from importlib import reload, invalidate_caches
|
|
19
19
|
import pyximport
|
|
20
20
|
from pkg_resources import get_distribution
|
|
21
21
|
|
|
@@ -67,8 +67,10 @@ cdef class Evaluation :
|
|
|
67
67
|
cdef public DICTIONARY
|
|
68
68
|
cdef public THREADS
|
|
69
69
|
cdef public A
|
|
70
|
+
cdef public regularisation_params
|
|
70
71
|
cdef public x
|
|
71
72
|
cdef public CONFIG
|
|
73
|
+
cdef public temp_data
|
|
72
74
|
cdef public confidence_map_img
|
|
73
75
|
|
|
74
76
|
def __init__( self, study_path='.', subject='.' ) :
|
|
@@ -81,18 +83,20 @@ cdef class Evaluation :
|
|
|
81
83
|
subject : string
|
|
82
84
|
The path (relative to previous folder) to the subject folder (default : '.')
|
|
83
85
|
"""
|
|
84
|
-
self.niiDWI
|
|
85
|
-
self.scheme
|
|
86
|
-
self.model
|
|
87
|
-
self.KERNELS
|
|
88
|
-
self.DICTIONARY
|
|
89
|
-
self.THREADS
|
|
90
|
-
self.A
|
|
91
|
-
self.
|
|
92
|
-
self.
|
|
86
|
+
self.niiDWI = None # set by "load_data" method
|
|
87
|
+
self.scheme = None # set by "load_data" method
|
|
88
|
+
self.model = None # set by "set_model" method
|
|
89
|
+
self.KERNELS = None # set by "load_kernels" method
|
|
90
|
+
self.DICTIONARY = None # set by "load_dictionary" method
|
|
91
|
+
self.THREADS = None # set by "set_threads" method
|
|
92
|
+
self.A = None # set by "build_operator" method
|
|
93
|
+
self.regularisation_params = None # set by "set_regularisation" method
|
|
94
|
+
self.x = None # set by "fit" method
|
|
95
|
+
self.confidence_map_img = None # set by "fit" method
|
|
93
96
|
|
|
94
97
|
# store all the parameters of an evaluation with COMMIT
|
|
95
98
|
self.CONFIG = {}
|
|
99
|
+
self.temp_data = {}
|
|
96
100
|
self.set_config('version', get_distribution('dmri-commit').version)
|
|
97
101
|
self.set_config('study_path', study_path)
|
|
98
102
|
self.set_config('subject', subject)
|
|
@@ -107,13 +111,18 @@ cdef class Evaluation :
|
|
|
107
111
|
|
|
108
112
|
def set_config( self, key, value ) :
|
|
109
113
|
self.CONFIG[ key ] = value
|
|
114
|
+
self.temp_data[ key ] = value
|
|
110
115
|
|
|
111
116
|
|
|
112
117
|
def get_config( self, key ) :
|
|
113
118
|
return self.CONFIG.get( key )
|
|
114
119
|
|
|
115
120
|
|
|
116
|
-
def
|
|
121
|
+
def get_temp_data( self, key ) :
|
|
122
|
+
return self.temp_data.get( key )
|
|
123
|
+
|
|
124
|
+
|
|
125
|
+
def load_data( self, dwi_filename, scheme_filename=None, b0_thr=0, b0_min_signal=0, replace_bad_voxels=None ) :
|
|
117
126
|
"""Load the diffusion signal and its corresponding acquisition scheme.
|
|
118
127
|
|
|
119
128
|
Parameters
|
|
@@ -170,7 +179,7 @@ cdef class Evaluation :
|
|
|
170
179
|
ERROR( 'Scheme does not match with input data' )
|
|
171
180
|
if self.scheme.dwi_count == 0 :
|
|
172
181
|
ERROR( 'There are no DWI volumes in the data' )
|
|
173
|
-
|
|
182
|
+
|
|
174
183
|
# Check for Nan or Inf values in raw data
|
|
175
184
|
if np.isnan(self.niiDWI_img).any() or np.isinf(self.niiDWI_img).any():
|
|
176
185
|
if replace_bad_voxels is not None:
|
|
@@ -240,7 +249,7 @@ cdef class Evaluation :
|
|
|
240
249
|
"""
|
|
241
250
|
# Call the specific model constructor
|
|
242
251
|
if hasattr(commit.models, model_name ) :
|
|
243
|
-
self.model = getattr(commit.models,model_name)()
|
|
252
|
+
self.model = getattr(commit.models, model_name)()
|
|
244
253
|
else :
|
|
245
254
|
ERROR( 'Model "%s" not recognized' % model_name )
|
|
246
255
|
|
|
@@ -271,7 +280,11 @@ cdef class Evaluation :
|
|
|
271
280
|
if self.model.id=='VolumeFractions' and ndirs!=1:
|
|
272
281
|
ndirs = 1
|
|
273
282
|
print( '\t* Forcing "ndirs" to 1 because model is isotropic' )
|
|
274
|
-
|
|
283
|
+
if 'commitwipmodels' in sys.modules :
|
|
284
|
+
if self.model.restrictedISO is not None and ndirs!=1:
|
|
285
|
+
ndirs = 1
|
|
286
|
+
print( '\t* Forcing "ndirs" to 1 because model is isotropic' )
|
|
287
|
+
|
|
275
288
|
# store some values for later use
|
|
276
289
|
self.set_config('lmax', lmax)
|
|
277
290
|
self.set_config('ndirs', ndirs)
|
|
@@ -482,21 +495,20 @@ cdef class Evaluation :
|
|
|
482
495
|
|
|
483
496
|
self.DICTIONARY['ISO'] = {}
|
|
484
497
|
|
|
485
|
-
self.DICTIONARY['
|
|
498
|
+
self.DICTIONARY['ISO']['v'] = np.fromfile( pjoin(self.get_config('TRACKING_path'),'dictionary_ISO_v.dict'), dtype=np.uint32 )
|
|
499
|
+
if self.DICTIONARY['ISO']['v'].size > 0:
|
|
500
|
+
self.DICTIONARY['ISO']['nV'] = self.DICTIONARY['ISO']['v'].size
|
|
501
|
+
else:
|
|
502
|
+
self.DICTIONARY['ISO']['nV'] = self.DICTIONARY['MASK'].sum()
|
|
486
503
|
|
|
487
|
-
|
|
488
|
-
vx = vx.astype(np.int32)
|
|
489
|
-
vy = vy.astype(np.int32)
|
|
490
|
-
vz = vz.astype(np.int32)
|
|
491
|
-
self.DICTIONARY['ISO']['v'] = vx + self.get_config('dim')[0] * ( vy + self.get_config('dim')[1] * vz )
|
|
492
|
-
del vx, vy, vz
|
|
504
|
+
self.DICTIONARY['nV'] = self.DICTIONARY['MASK'].sum()
|
|
493
505
|
|
|
494
506
|
# reorder the segments based on the "v" field
|
|
495
507
|
idx = np.argsort( self.DICTIONARY['ISO']['v'], kind='mergesort' )
|
|
496
508
|
self.DICTIONARY['ISO']['v'] = self.DICTIONARY['ISO']['v'][ idx ]
|
|
497
509
|
del idx
|
|
498
510
|
|
|
499
|
-
print( '[ %d voxels ]' % self.DICTIONARY['nV'] )
|
|
511
|
+
print( '[ %d voxels ]' % self.DICTIONARY['ISO']['nV'] )
|
|
500
512
|
|
|
501
513
|
# post-processing
|
|
502
514
|
# ---------------
|
|
@@ -591,10 +603,24 @@ cdef class Evaluation :
|
|
|
591
603
|
self.THREADS['EC'] = None
|
|
592
604
|
|
|
593
605
|
if self.DICTIONARY['nV'] > 0 :
|
|
594
|
-
self.
|
|
595
|
-
|
|
596
|
-
|
|
597
|
-
|
|
606
|
+
if self.DICTIONARY['ISO']['nV'] > 0 :
|
|
607
|
+
self.THREADS['ISO'] = np.zeros( n+1, dtype=np.uint32 )
|
|
608
|
+
N = np.floor( self.DICTIONARY['ISO']['nV']/n )
|
|
609
|
+
t = 1
|
|
610
|
+
tot = 0
|
|
611
|
+
C = np.bincount( self.DICTIONARY['ISO']['v'] )
|
|
612
|
+
for c in C :
|
|
613
|
+
tot += c
|
|
614
|
+
if tot >= N and t <= n :
|
|
615
|
+
self.THREADS['ISO'][t] = self.THREADS['ISO'][t-1] + tot
|
|
616
|
+
t += 1
|
|
617
|
+
tot = 0
|
|
618
|
+
self.THREADS['ISO'][n] = self.DICTIONARY['ISO']['nV']
|
|
619
|
+
else:
|
|
620
|
+
self.THREADS['ISO'] = np.zeros( n+1, dtype=np.uint32 )
|
|
621
|
+
for i in xrange(n) :
|
|
622
|
+
self.THREADS['ISO'][i] = np.searchsorted( self.DICTIONARY['ISO']['v'], self.DICTIONARY['IC']['v'][ self.THREADS['IC'][i] ] )
|
|
623
|
+
self.THREADS['ISO'][n] = self.DICTIONARY['ISO']['nV']
|
|
598
624
|
|
|
599
625
|
# check if some threads are not assigned any segment
|
|
600
626
|
if np.count_nonzero( np.diff( self.THREADS['ISO'].astype(np.int32) ) <= 0 ) :
|
|
@@ -647,10 +673,11 @@ cdef class Evaluation :
|
|
|
647
673
|
|
|
648
674
|
if self.DICTIONARY['nV'] > 0 :
|
|
649
675
|
self.THREADS['ISOt'] = np.zeros( n+1, dtype=np.uint32 )
|
|
650
|
-
N = np.floor( self.DICTIONARY['nV']/n )
|
|
676
|
+
N = np.floor( self.DICTIONARY['ISO']['nV']/n )
|
|
677
|
+
|
|
651
678
|
for i in xrange(1,n) :
|
|
652
679
|
self.THREADS['ISOt'][i] = self.THREADS['ISOt'][i-1] + N
|
|
653
|
-
self.THREADS['ISOt'][n] = self.DICTIONARY['nV']
|
|
680
|
+
self.THREADS['ISOt'][n] = self.DICTIONARY['ISO']['nV']
|
|
654
681
|
|
|
655
682
|
# check if some threads are not assigned any segment
|
|
656
683
|
if np.count_nonzero( np.diff( self.THREADS['ISOt'].astype(np.int32) ) <= 0 ) :
|
|
@@ -728,14 +755,14 @@ cdef class Evaluation :
|
|
|
728
755
|
config.nISO = self.KERNELS['iso'].shape[0]
|
|
729
756
|
config.build_dir = build_dir
|
|
730
757
|
|
|
758
|
+
sys.dont_write_bytecode = True
|
|
731
759
|
pyximport.install( reload_support=True, language_level=3, build_dir=build_dir, build_in_temp=True, inplace=False )
|
|
732
760
|
|
|
733
|
-
|
|
734
|
-
|
|
735
|
-
|
|
736
|
-
reload( sys.modules['commit.operator.operator'] )
|
|
761
|
+
if 'commit.operator.operator' in sys.modules :
|
|
762
|
+
del sys.modules['commit.operator.operator']
|
|
763
|
+
import commit.operator.operator
|
|
737
764
|
|
|
738
|
-
self.A =
|
|
765
|
+
self.A = commit.operator.operator.LinearOperator( self.DICTIONARY, self.KERNELS, self.THREADS )
|
|
739
766
|
|
|
740
767
|
LOG( ' [ %.1f seconds ]' % ( time.time() - tic ) )
|
|
741
768
|
|
|
@@ -751,11 +778,182 @@ cdef class Evaluation :
|
|
|
751
778
|
ERROR( 'Data not loaded; call "load_data()" first' )
|
|
752
779
|
|
|
753
780
|
y = self.niiDWI_img[ self.DICTIONARY['MASK_ix'], self.DICTIONARY['MASK_iy'], self.DICTIONARY['MASK_iz'], : ].flatten().astype(np.float64)
|
|
754
|
-
y[y < 0] = 0
|
|
781
|
+
# y[y < 0] = 0
|
|
755
782
|
return y
|
|
756
783
|
|
|
784
|
+
def set_regularisation(self, regularisers=(None, None, None), lambdas=(None, None, None), is_nonnegative=(True, True, True), params=(None, None, None)):
|
|
785
|
+
"""
|
|
786
|
+
Set the regularisation parameters for the optimisation problem.
|
|
787
|
+
Input
|
|
788
|
+
-----
|
|
789
|
+
regularisers - tuple :
|
|
790
|
+
sets the penalty term to be used for each compartment:
|
|
791
|
+
regularisers[0] corresponds to the Intracellular compartment
|
|
792
|
+
regularisers[1] corresponds to the Extracellular compartment
|
|
793
|
+
regularisers[2] corresponds to the Isotropic compartment
|
|
794
|
+
Each regularisers[k] must be one of: {None, 'lasso', weighted_lasso, 'group_lasso'}:
|
|
795
|
+
'lasso' penalises with the 1-norm of the coefficients
|
|
796
|
+
corresponding to the compartment.
|
|
797
|
+
'group_lasso' penalises according to the following formulation (see [1]):
|
|
798
|
+
$\Omega(x) = \lambda \sum_{g\in G} w_g |x_g|
|
|
799
|
+
Considers both the non-overlapping and the hierarchical formulations.
|
|
800
|
+
NB: this option is allowed only in the IC compartment.
|
|
801
|
+
Default = (None, None, None).
|
|
802
|
+
|
|
803
|
+
lambdas - tuple :
|
|
804
|
+
regularisation parameter for each compartment.
|
|
805
|
+
The lambdas correspond to the ones described in the mathematical
|
|
806
|
+
formulation of the regularisation term
|
|
807
|
+
$\Omega(x) = lambdas[0]*regnorm[0](x) + lambdas[1]*regnorm[1](x) + lambdas[2]*regnorm[2](x)$
|
|
808
|
+
Default = (0.0, 0.0, 0.0, 0.0).
|
|
809
|
+
|
|
810
|
+
is_nonnegative - tuple :
|
|
811
|
+
impose a non negativity constraint for each compartment:
|
|
812
|
+
is_nonnegative[0] corresponds to the Intracellular compartment
|
|
813
|
+
is_nonnegative[1] corresponds to the Extracellular compartment
|
|
814
|
+
is_nonnegative[2] corresponds to the Isotropic compartment
|
|
815
|
+
Default = (True, True, True).
|
|
816
|
+
|
|
817
|
+
structureIC - np.array(list(list), dtype=np.object_) :
|
|
818
|
+
group structure for the IC compartment.
|
|
819
|
+
This field is necessary only if regterm[0]='group_lasso'.
|
|
820
|
+
Example:
|
|
821
|
+
structureIC = np.array([[0,2,5],[1,3,4],[0,1,2,3,4,5],[6]], dtype=np.object_)
|
|
822
|
+
|
|
823
|
+
that is equivalent to
|
|
824
|
+
[0,1,2,3,4,5] [6]
|
|
825
|
+
/ \
|
|
826
|
+
[0,2,5] [1,3,4]
|
|
827
|
+
which has two non overlapping groups, one of which is the union
|
|
828
|
+
of two other non-overlapping groups.
|
|
829
|
+
|
|
830
|
+
weightsIC - np.array(np.float64) :
|
|
831
|
+
this defines the weights associated to each element of structure IC.
|
|
832
|
+
|
|
833
|
+
weightsIC_group - np.array(np.float64) :
|
|
834
|
+
this defines the weights associated to each group of structure IC.
|
|
835
|
+
|
|
836
|
+
"""
|
|
837
|
+
|
|
838
|
+
regularisation = {}
|
|
839
|
+
|
|
840
|
+
regularisation['startIC'] = 0
|
|
841
|
+
regularisation['sizeIC'] = int( self.DICTIONARY['IC']['nF'] * self.KERNELS['wmr'].shape[0])
|
|
842
|
+
regularisation['startEC'] = int( regularisation['sizeIC'] )
|
|
843
|
+
regularisation['sizeEC'] = int( self.DICTIONARY['EC']['nE'] * self.KERNELS['wmh'].shape[0])
|
|
844
|
+
regularisation['startISO'] = int( regularisation['sizeIC'] + regularisation['sizeEC'] )
|
|
845
|
+
regularisation['sizeISO'] = int( self.DICTIONARY['nV'] * self.KERNELS['iso'].shape[0])
|
|
846
|
+
|
|
847
|
+
regularisation['regIC'] = regularisers[0]
|
|
848
|
+
regularisation['regEC'] = regularisers[1]
|
|
849
|
+
regularisation['regISO'] = regularisers[2]
|
|
850
|
+
|
|
851
|
+
regularisation['nnIC'] = is_nonnegative[0]
|
|
852
|
+
regularisation['nnEC'] = is_nonnegative[1]
|
|
853
|
+
regularisation['nnISO'] = is_nonnegative[2]
|
|
854
|
+
|
|
855
|
+
regularisation['paramsIC'] = params[0]
|
|
856
|
+
regularisation['paramsEC'] = params[1]
|
|
857
|
+
regularisation['paramsISO'] = params[2]
|
|
757
858
|
|
|
758
|
-
|
|
859
|
+
regularisation['dictIC_params'] = params[0]
|
|
860
|
+
regularisation['dictEC_params'] = params[1]
|
|
861
|
+
regularisation['dictISO_params'] = params[2]
|
|
862
|
+
|
|
863
|
+
# unpack parameters inside params as three separate dictionaries
|
|
864
|
+
dictIC_params, dictEC_params, dictISO_params = params
|
|
865
|
+
|
|
866
|
+
if regularisation['regIC'] == 'lasso':
|
|
867
|
+
if lambdas[0] is None:
|
|
868
|
+
raise ValueError('Missing regularisation parameter for the IC compartment')
|
|
869
|
+
else:
|
|
870
|
+
regularisation['lambdaIC'] = lambdas[0]
|
|
871
|
+
elif regularisation['regEC'] == 'lasso':
|
|
872
|
+
if lambdas[1] is None:
|
|
873
|
+
raise ValueError('Missing regularisation parameter for the EC compartment')
|
|
874
|
+
else:
|
|
875
|
+
regularisation['lambdaEC'] = lambdas[1]
|
|
876
|
+
elif regularisation['regISO'] == 'lasso':
|
|
877
|
+
if lambdas[2] is None:
|
|
878
|
+
raise ValueError('Missing regularisation parameter for the ISO compartment')
|
|
879
|
+
else:
|
|
880
|
+
regularisation['lambdaISO'] = lambdas[2]
|
|
881
|
+
|
|
882
|
+
elif regularisation['regIC'] == 'smoothness':
|
|
883
|
+
if lambdas[0] is None:
|
|
884
|
+
raise ValueError('Missing regularisation parameter for the IC compartment')
|
|
885
|
+
else:
|
|
886
|
+
regularisation['lambdaIC'] = lambdas[0]
|
|
887
|
+
elif regularisation['regEC'] == 'smoothness':
|
|
888
|
+
raise ValueError('Not yet implemented')
|
|
889
|
+
elif regularisation['regISO'] == 'smoothness':
|
|
890
|
+
raise ValueError('Not yet implemented')
|
|
891
|
+
|
|
892
|
+
elif regularisation['regIC'] == 'group_lasso':
|
|
893
|
+
if lambdas[0] is None:
|
|
894
|
+
raise ValueError('Group structure for the IC compartment not provided')
|
|
895
|
+
else:
|
|
896
|
+
regularisation['lambdaIC'] = lambdas[0]
|
|
897
|
+
if lambdas[1] is not None:
|
|
898
|
+
raise ValueError('Not yet implemented')
|
|
899
|
+
if lambdas[2] is not None:
|
|
900
|
+
raise ValueError('Not yet implemented')
|
|
901
|
+
if dictIC_params["group_idx"] is None:
|
|
902
|
+
raise ValueError('Group structure for the IC compartment not provided')
|
|
903
|
+
if dictIC_params["group_weights"] is None:
|
|
904
|
+
raise ValueError('Group weights for the IC compartment not provided')
|
|
905
|
+
elif regularisation['regEC'] == 'group_lasso':
|
|
906
|
+
raise ValueError('Not yet implemented')
|
|
907
|
+
elif regularisation['regISO'] == 'group_lasso':
|
|
908
|
+
raise ValueError('Not yet implemented')
|
|
909
|
+
|
|
910
|
+
elif regularisation['regIC'] == 'sparse_group_lasso':
|
|
911
|
+
if len(lambdas[0]) != 2:
|
|
912
|
+
raise ValueError('Group structure for the IC compartment not provided')
|
|
913
|
+
else:
|
|
914
|
+
regularisation['lambdaIC'] = lambdas[0]
|
|
915
|
+
if lambdas[1] is not None:
|
|
916
|
+
raise ValueError('Not yet implemented')
|
|
917
|
+
if lambdas[2] is not None:
|
|
918
|
+
raise ValueError('Not yet implemented')
|
|
919
|
+
if dictIC_params["group_idx"] is None:
|
|
920
|
+
raise ValueError('Group structure for the IC compartment not provided')
|
|
921
|
+
if dictIC_params["group_weights"] is None:
|
|
922
|
+
raise ValueError('Group weights for the IC compartment not provided')
|
|
923
|
+
elif regularisation['regEC'] == 'sparse_group_lasso':
|
|
924
|
+
raise ValueError('Not yet implemented')
|
|
925
|
+
elif regularisation['regISO'] == 'sparse_group_lasso':
|
|
926
|
+
raise ValueError('Not yet implemented')
|
|
927
|
+
|
|
928
|
+
# Check if group indices need to be updated in case of 'group_lasso' or 'sparse_group_lasso'
|
|
929
|
+
if regularisation['regIC'] == 'group_lasso' or regularisation['regIC'] == 'sparse_group_lasso' and (0 in self.DICTIONARY['TRK']['kept']) :
|
|
930
|
+
dictionary_TRK_kept = self.DICTIONARY['TRK']['kept']
|
|
931
|
+
weightsIC_group = dictIC_params["group_weights"]
|
|
932
|
+
idx_in_kept = np.zeros(dictionary_TRK_kept.size, dtype=np.int32) - 1 # -1 is used to flag indices for removal
|
|
933
|
+
idx_in_kept[dictionary_TRK_kept==1] = list(range(self.DICTIONARY['IC']['nF']))
|
|
934
|
+
|
|
935
|
+
newICgroup_idx = []
|
|
936
|
+
newweightsIC_group = []
|
|
937
|
+
ICgroup_idx = dictIC_params["group_idx"]
|
|
938
|
+
for count, group in enumerate(ICgroup_idx):
|
|
939
|
+
group = idx_in_kept[group]
|
|
940
|
+
idx_to_delete = np.where(group==-1)[0]
|
|
941
|
+
if idx_to_delete.size>0:
|
|
942
|
+
group = np.delete(group,idx_to_delete)
|
|
943
|
+
if(group.size>0):
|
|
944
|
+
newICgroup_idx.append(group)
|
|
945
|
+
newweightsIC_group.append(weightsIC_group[count])
|
|
946
|
+
else:
|
|
947
|
+
newICgroup_idx.append(group)
|
|
948
|
+
newweightsIC_group.append(weightsIC_group[count])
|
|
949
|
+
|
|
950
|
+
dictIC_params["group_idx"] = np.array(newICgroup_idx, dtype=np.object_)
|
|
951
|
+
dictIC_params['group_weights'] = np.array(newweightsIC_group)
|
|
952
|
+
|
|
953
|
+
self.regularisation_params = commit.solvers.init_regularisation(regularisation)
|
|
954
|
+
|
|
955
|
+
|
|
956
|
+
def fit( self, tol_fun=1e-3, tol_x=1e-6, max_iter=100, verbose=True, x0=None, confidence_map_filename=None, confidence_map_rescale=False ) :
|
|
759
957
|
"""Fit the model to the data.
|
|
760
958
|
|
|
761
959
|
Parameters
|
|
@@ -765,14 +963,9 @@ cdef class Evaluation :
|
|
|
765
963
|
max_iter : integer
|
|
766
964
|
Maximum number of iterations (default : 100)
|
|
767
965
|
verbose : boolean
|
|
768
|
-
Level of verbosity: 0=no print, 1=print progress (default :
|
|
966
|
+
Level of verbosity: 0=no print, 1=print info at each iteration, 2=print progress bar, 3=print debug info (default : 2)
|
|
769
967
|
x0 : np.array
|
|
770
968
|
Initial guess for the solution of the problem (default : None)
|
|
771
|
-
regularisation : commit.solvers.init_regularisation object
|
|
772
|
-
Python dictionary that describes the wanted regularisation term.
|
|
773
|
-
Check the documentation of commit.solvers.init_regularisation to see
|
|
774
|
-
how to properly define the wanted mathematical formulation
|
|
775
|
-
(default : None)
|
|
776
969
|
confidence_map_filename : string
|
|
777
970
|
Path to the NIFTI file containing a confidence map on the data,
|
|
778
971
|
relative to the subject folder. The file can be 3D or 4D in
|
|
@@ -864,21 +1057,23 @@ cdef class Evaluation :
|
|
|
864
1057
|
if x0 is not None :
|
|
865
1058
|
if x0.shape[0] != self.A.shape[1] :
|
|
866
1059
|
ERROR( 'x0 dimension does not match the number of columns of the dictionary' )
|
|
867
|
-
|
|
868
|
-
|
|
1060
|
+
|
|
1061
|
+
if self.regularisation_params is None:
|
|
1062
|
+
self.set_regularisation()
|
|
1063
|
+
|
|
869
1064
|
|
|
870
1065
|
self.CONFIG['optimization'] = {}
|
|
871
1066
|
self.CONFIG['optimization']['tol_fun'] = tol_fun
|
|
872
1067
|
self.CONFIG['optimization']['tol_x'] = tol_x
|
|
873
1068
|
self.CONFIG['optimization']['max_iter'] = max_iter
|
|
874
1069
|
self.CONFIG['optimization']['verbose'] = verbose
|
|
875
|
-
self.CONFIG['optimization']['regularisation'] =
|
|
1070
|
+
self.CONFIG['optimization']['regularisation'] = self.regularisation_params
|
|
876
1071
|
|
|
877
1072
|
# run solver
|
|
878
1073
|
t = time.time()
|
|
879
1074
|
LOG( '\n-> Fit model:' )
|
|
880
1075
|
|
|
881
|
-
self.x, opt_details = commit.solvers.solve(self.get_y(), self.A, self.A.T, tol_fun
|
|
1076
|
+
self.x, opt_details = commit.solvers.solve(self.get_y(), self.A, self.A.T, tol_fun=tol_fun, tol_x=tol_x, max_iter=max_iter, verbose=verbose, x0=x0, regularisation=self.regularisation_params, confidence_array=confidence_array)
|
|
882
1077
|
|
|
883
1078
|
self.CONFIG['optimization']['fit_details'] = opt_details
|
|
884
1079
|
self.CONFIG['optimization']['fit_time'] = time.time()-t
|
|
@@ -1071,7 +1266,9 @@ cdef class Evaluation :
|
|
|
1071
1266
|
niiISO_img = np.zeros( self.get_config('dim'), dtype=np.float32 )
|
|
1072
1267
|
if len(self.KERNELS['iso']) > 0 :
|
|
1073
1268
|
offset = nF * self.KERNELS['wmr'].shape[0] + nE * self.KERNELS['wmh'].shape[0]
|
|
1074
|
-
|
|
1269
|
+
offset_iso = offset + self.DICTIONARY['ISO']['nV']
|
|
1270
|
+
tmp = x[offset:offset_iso].reshape( (-1,self.DICTIONARY['ISO']['nV']) ).sum( axis=0 )
|
|
1271
|
+
xv = np.bincount( self.DICTIONARY['ISO']['v'], weights=tmp, minlength=nV ).astype(np.float32)
|
|
1075
1272
|
niiISO_img[ self.DICTIONARY['MASK_ix'], self.DICTIONARY['MASK_iy'], self.DICTIONARY['MASK_iz'] ] = xv
|
|
1076
1273
|
print( ' [ OK ]' )
|
|
1077
1274
|
|
|
@@ -1091,8 +1288,6 @@ cdef class Evaluation :
|
|
|
1091
1288
|
# Configuration and results
|
|
1092
1289
|
print( '\t* Configuration and results:' )
|
|
1093
1290
|
|
|
1094
|
-
print( '\t\t- streamline_weights.txt... ', end='' )
|
|
1095
|
-
sys.stdout.flush()
|
|
1096
1291
|
xic, _, _ = self.get_coeffs()
|
|
1097
1292
|
if stat_coeffs != 'all' and xic.size > 0 :
|
|
1098
1293
|
xic = np.reshape( xic, (-1,self.DICTIONARY['TRK']['kept'].size) )
|
|
@@ -1118,7 +1313,7 @@ cdef class Evaluation :
|
|
|
1118
1313
|
ordered_idx = dictionary_info["tractogram_centr_idx"].astype(np.int64)
|
|
1119
1314
|
unravel_weights = np.zeros( dictionary_info['n_count'], dtype=np.float64)
|
|
1120
1315
|
unravel_weights[ordered_idx] = self.DICTIONARY['TRK']['kept'].astype(np.float64)
|
|
1121
|
-
temp_weights = unravel_weights[ordered_idx]
|
|
1316
|
+
temp_weights = unravel_weights[ordered_idx]
|
|
1122
1317
|
if dictionary_info['blur_gauss_extent'] > 0 or dictionary_info['blur_core_extent'] > 0:
|
|
1123
1318
|
temp_weights[temp_weights>0] = xic[self.DICTIONARY['TRK']['kept']>0] * self.DICTIONARY['TRK']['lenTot'] / self.DICTIONARY['TRK']['len']
|
|
1124
1319
|
unravel_weights[ordered_idx] = temp_weights
|
|
@@ -1127,11 +1322,26 @@ cdef class Evaluation :
|
|
|
1127
1322
|
temp_weights[temp_weights>0] = xic[self.DICTIONARY['TRK']['kept']>0]
|
|
1128
1323
|
unravel_weights[ordered_idx] = temp_weights
|
|
1129
1324
|
xic = unravel_weights
|
|
1130
|
-
|
|
1325
|
+
|
|
1131
1326
|
else:
|
|
1132
1327
|
if dictionary_info['blur_gauss_extent'] > 0 or dictionary_info['blur_core_extent'] > 0:
|
|
1133
1328
|
xic[ self.DICTIONARY['TRK']['kept']==1 ] *= self.DICTIONARY['TRK']['lenTot'] / self.DICTIONARY['TRK']['len']
|
|
1134
1329
|
|
|
1330
|
+
|
|
1331
|
+
self.temp_data['DICTIONARY'] = self.DICTIONARY
|
|
1332
|
+
self.temp_data['niiIC_img'] = niiIC_img
|
|
1333
|
+
self.temp_data['niiEC_img'] = niiEC_img
|
|
1334
|
+
self.temp_data['niiISO_img'] = niiISO_img
|
|
1335
|
+
self.temp_data['streamline_weights'] = xic
|
|
1336
|
+
self.temp_data['RESULTS_path'] = RESULTS_path
|
|
1337
|
+
|
|
1338
|
+
if hasattr(self.model, '_postprocess'):
|
|
1339
|
+
self.model._postprocess(self.temp_data, verbose=self.CONFIG['optimization']['verbose'])
|
|
1340
|
+
else:
|
|
1341
|
+
print( '\t\t- streamline_weights.txt... ', end='' )
|
|
1342
|
+
sys.stdout.flush()
|
|
1343
|
+
|
|
1344
|
+
|
|
1135
1345
|
np.savetxt( pjoin(RESULTS_path,'streamline_weights.txt'), xic, fmt=coeffs_format )
|
|
1136
1346
|
self.set_config('stat_coeffs', stat_coeffs)
|
|
1137
1347
|
print( '[ OK ]' )
|
|
@@ -1142,6 +1352,13 @@ cdef class Evaluation :
|
|
|
1142
1352
|
# item 2: np.array renormalisation of coeffs in item 1
|
|
1143
1353
|
print( '\t\t- results.pickle... ', end='' )
|
|
1144
1354
|
sys.stdout.flush()
|
|
1355
|
+
|
|
1356
|
+
if self.CONFIG['optimization']['regularisation'] is not None:
|
|
1357
|
+
if 'omega' in self.CONFIG['optimization']['regularisation']:
|
|
1358
|
+
self.CONFIG['optimization']['regularisation'].pop('omega')
|
|
1359
|
+
if 'prox' in self.CONFIG['optimization']['regularisation']:
|
|
1360
|
+
self.CONFIG['optimization']['regularisation'].pop('prox')
|
|
1361
|
+
|
|
1145
1362
|
with open( pjoin(RESULTS_path,'results.pickle'), 'wb+' ) as fid :
|
|
1146
1363
|
pickle.dump( [self.CONFIG, self.x, x], fid, protocol=2 )
|
|
1147
1364
|
print( ' [ OK ]' )
|
|
@@ -2,7 +2,7 @@
|
|
|
2
2
|
|
|
3
3
|
# Format version as expected by setup.py (string of form "X.Y.Z")
|
|
4
4
|
_version_major = 2
|
|
5
|
-
_version_minor =
|
|
5
|
+
_version_minor = 1
|
|
6
6
|
_version_micro = 0
|
|
7
7
|
_version_extra = '' #'.dev'
|
|
8
8
|
__version__ = "%s.%s.%s%s" % (_version_major,_version_minor,_version_micro,_version_extra)
|
|
@@ -23,7 +23,7 @@ These weights can be easily estimated by solving a convenient global convex opti
|
|
|
23
23
|
"""
|
|
24
24
|
URL = 'https://github.com/daducci/COMMIT'
|
|
25
25
|
DOWNLOAD_URL = "N/A"
|
|
26
|
-
LICENSE = '
|
|
26
|
+
LICENSE = 'Proprietary license'
|
|
27
27
|
AUTHOR = 'Alessandro Daducci'
|
|
28
28
|
AUTHOR_EMAIL = 'alessandro.daducci@univr.it'
|
|
29
29
|
PLATFORMS = "OS independent"
|