Add gradient and autodiff checks for linear operators

[ckolbPTB]

This supersedes #407 - it was too difficult to bring this up-to-date with main.

forward_mode_autodiff_of_linear_operator_test and gradient_of_linear_operator_test still need to be added to

• ZeroOp
• RearrangeOp
• PCACompOp
• GridSamplingOp

Do we want to run forward_mode_autodiff_of_linear_operator_test and gradient_of_linear_operator_test on all parametrize-options used for the adjointness tests?

[github-actions]

Coverage Report

File	Stmts	Miss	Cover	Missing
src/mrpro/algorithms/csm
inati.py	24	1	96%	44
walsh.py	16	1	94%	53
src/mrpro/algorithms/dcf
dcf_voronoi.py	53	4	92%	15, 55–56, 89
src/mrpro/algorithms/optimizers
adam.py	20	1	95%	101
pdhg.py	79	3	96%	177–178, 184
src/mrpro/algorithms/reconstruction
DirectReconstruction.py	28	16	43%	59–79, 93
IterativeSENSEReconstruction.py	13	1	92%	79
Reconstruction.py	50	22	56%	42, 54–56, 80–87, 108–117
RegularizedIterativeSENSEReconstruction.py	41	17	59%	97–101, 115–140
src/mrpro/data
AcqInfo.py	128	3	98%	26, 169, 207
CsmData.py	29	3	90%	15, 84–86
DcfData.py	45	8	82%	18, 66, 78–83
IData.py	67	9	87%	120, 126, 130, 160–168
IHeader.py	75	7	91%	75, 109, 127–131
KData.py	214	25	88%	111–112, 127, 134, 144, 152, 206–207, 245, 250–251, 270–281, 440, 442, 507, 522, 559, 590, 599
KHeader.py	153	17	89%	25, 119–123, 150, 199, 210, 217–218, 221, 228, 260–271
KNoise.py	31	15	52%	39–52, 56–61
KTrajectory.py	81	12	85%	112–117, 120–122, 207–211
MoveDataMixin.py	140	18	87%	28, 126, 142, 156–158, 220, 336–338, 351, 430, 450–451, 453, 468–469, 471
QData.py	39	7	82%	42, 65–73
Rotation.py	674	35	95%	100, 198, 335, 433, 477, 495, 582, 584, 593, 627, 629, 692, 769, 774, 777, 792, 809, 814, 890, 1078, 1083, 1086, 1110, 1114, 1242, 1244, 1252–1253, 1317, 1399, 1692, 1844, 1879, 1883, 1994
SpatialDimension.py	232	21	91%	34, 104, 141, 148, 154, 274–276, 289–291, 325, 343, 356, 369, 382, 395, 404–405, 420, 429
acq_filters.py	12	1	92%	47
src/mrpro/data/traj_calculators
KTrajectoryCalculator.py	25	2	92%	23, 45
KTrajectoryIsmrmrd.py	13	2	85%	41, 50
KTrajectoryPulseq.py	23	1	96%	55
src/mrpro/operators
CartesianSamplingOp.py	89	3	97%	118, 157, 280
ConstraintsOp.py	60	2	97%	46, 48
EndomorphOperator.py	65	2	97%	228, 234
FiniteDifferenceOp.py	27	2	93%	40, 105
FourierOp.py	158	3	98%	274, 392, 397
Functional.py	71	5	93%	20–22, 117, 119
GridSamplingOp.py	136	9	93%	72–73, 82–83, 90–91, 94, 96, 98
LinearOperator.py	168	8	95%	114, 224, 254, 295, 304, 312, 329, 364
LinearOperatorMatrix.py	158	16	90%	82, 119, 152, 161, 166, 175–178, 191–194, 203, 215, 304, 331, 359
MultiIdentityOp.py	13	2	85%	43, 48
Operator.py	78	2	97%	31, 87
ProximableFunctionalSeparableSum.py	39	3	92%	50, 103, 110
SliceProjectionOp.py	173	8	95%	44, 61, 63, 69, 207, 228, 261, 301
WaveletOp.py	119	5	96%	151, 169, 204, 209, 232
ZeroPadOp.py	16	1	94%	30
src/mrpro/utils
filters.py	62	2	97%	44, 49
reshape.py	60	1	98%	191
slice_profiles.py	46	6	87%	20, 36, 113–116, 149
sliding_window.py	34	1	97%	34
split_idx.py	10	2	80%	43, 47
summarize_tensorvalues.py	11	9	18%	20–29
typing.py	21	11	48%	8–23
zero_pad_or_crop.py	31	6	81%	26, 30, 55, 58, 61, 64
TOTAL	4973	361	93%

Tests	Skipped	Failures	Errors	Time
2516	0 💤	0 ❌	0 🔥	1m 24s ⏱️

[github-actions]

📚 Documentation

📁 Download as zip
🔍 View online

[ckolbPTB]

Tests which currently do not pass:
FAILED tests/operators/test_slice_projection_op.py::test_slice_projection_op_grad - RuntimeError: Sparse CSR tensors do not have strides
FAILED tests/operators/test_slice_projection_op.py::test_slice_projection_op_forward_mode_autodiff - RuntimeError: Sparse CSR tensors do not have strides

[fzimmermann89]

That is a known issue #409 🙈
pytorch/pytorch#97286 pytorch/pytorch#136357

There is imho no feasibe way tfor us to support the slice projection operator in torch.func.

Fixing it in pytorch would also be infeasible for us, as the the functorch c backend seems to cause the issue, as sparse tensors or completely missing in its implementation,

We could replace the sparse-matrix/dense vector multiplication by manual indexing and dotproducts. This will make it quite a bit slower and way more complicated.

For now, I would ether remove the tests or mark them as failing.

This mainly means that we cant solve a superresolutionproblem with fista or pdgh if we keep using torch.func for the gradients.

[ckolbPTB]

@fzimmermann89 do you think you could add the missing tests for GridSamplingOp? Everthing else should now be covered...

[fzimmermann89]

I moved #604 out of this PR, to unblock #293

[github-actions]

Coverage Report

File	Stmts	Miss	Cover	Missing
src/mrpro
_version.py	6	2	67%	7–8
src/mrpro/algorithms/csm
inati.py	24	1	96%	44
walsh.py	16	1	94%	53
src/mrpro/algorithms/dcf
dcf_voronoi.py	55	4	93%	15, 55–56, 89
src/mrpro/algorithms/optimizers
adam.py	20	1	95%	101
pdhg.py	79	3	96%	177–178, 184
pgd.py	51	4	92%	106, 151–154
src/mrpro/algorithms/reconstruction
DirectReconstruction.py	28	16	43%	59–79, 93
IterativeSENSEReconstruction.py	13	1	92%	79
Reconstruction.py	50	22	56%	42, 54–56, 80–87, 108–117
RegularizedIterativeSENSEReconstruction.py	41	17	59%	97–101, 115–140
src/mrpro/data
AcqInfo.py	145	5	97%	42, 121–122, 124, 233
CsmData.py	29	3	90%	15, 84–86
DcfData.py	45	8	82%	18, 66, 78–83
EncodingLimits.py	73	3	96%	33, 123, 126
IData.py	59	8	86%	118, 132, 159–167
IHeader.py	136	12	91%	69–72, 253, 257, 261, 265, 299–303
KData.py	201	29	86%	113–114, 129, 136, 147–157, 166, 174, 183, 217, 239–241, 277–278, 333–344, 473, 475, 545
KHeader.py	141	21	85%	24, 109–115, 142, 190, 197–198, 201, 208, 225–232, 240–251
KNoise.py	31	15	52%	39–52, 56–61
KTrajectory.py	108	3	97%	166, 168, 188
MoveDataMixin.py	140	18	87%	28, 126, 142, 156–158, 220, 336–338, 351, 430, 450–451, 453, 468–469, 471
QData.py	39	7	82%	42, 65–73
Rotation.py	719	35	95%	101, 199, 336, 434, 478, 496, 583, 585, 594, 628, 630, 693, 770, 775, 778, 793, 810, 815, 891, 1079, 1084, 1087, 1111, 1115, 1243, 1245, 1253–1254, 1318, 1400, 1703, 1855, 1890, 1894, 2084
SpatialDimension.py	232	20	91%	34, 104, 148, 154, 274–276, 289–291, 325, 343, 356, 369, 382, 395, 404–405, 420, 429
acq_filters.py	12	1	92%	47
src/mrpro/data/traj_calculators
KTrajectoryCalculator.py	27	1	96%	84
KTrajectoryIsmrmrd.py	19	1	95%	57
src/mrpro/operators
CartesianSamplingOp.py	96	4	96%	124, 163, 238, 305
ConstraintsOp.py	60	2	97%	46, 48
EndomorphOperator.py	32	2	94%	52, 58
FiniteDifferenceOp.py	27	2	93%	40, 105
FourierOp.py	90	3	97%	185, 257, 262
Functional.py	77	9	88%	20–22, 117, 119, 226–228, 242
GridSamplingOp.py	136	9	93%	72–73, 82–83, 90–91, 94, 96, 98
LinearOperator.py	201	8	96%	107, 217, 244, 251, 292, 301, 309, 326
LinearOperatorMatrix.py	162	19	88%	82, 119, 152, 161, 166, 175–178, 191–194, 202–203, 208–209, 221, 310, 337, 364
MultiIdentityOp.py	13	2	85%	43, 48
NonUniformFastFourierOp.py	188	10	95%	69, 96, 206, 208, 241, 243, 319, 373, 423, 428
Operator.py	79	2	97%	32, 88
ProximableFunctionalSeparableSum.py	39	3	92%	50, 103, 110
SliceProjectionOp.py	174	8	95%	45, 62, 64, 70, 206, 227, 260, 300
WaveletOp.py	119	5	96%	151, 169, 204, 209, 232
ZeroPadOp.py	16	1	94%	30
src/mrpro/phantoms
brainweb.py	275	26	91%	325–335, 371, 427–430, 452–453, 458–459, 474, 520, 586–587, 606–609, 620, 622, 657–658, 671
src/mrpro/utils
filters.py	62	2	97%	44, 49
reshape.py	95	2	98%	110, 301
slice_profiles.py	47	6	87%	21, 37, 116–119, 152
sliding_window.py	34	1	97%	34
split_idx.py	10	2	80%	43, 47
summarize_tensorvalues.py	12	3	75%	23, 25, 27
typing.py	69	55	20%	9–235
unit_conversion.py	60	14	77%	32, 40, 42, 49, 51, 58, 60, 69, 80, 82, 101, 103, 124, 126
zero_pad_or_crop.py	31	6	81%	26, 30, 55, 58, 61, 64
TOTAL	5695	468	92%

Tests	Skipped	Failures	Errors	Time
2687	0 💤	0 ❌	0 🔥	2m 7s ⏱️

[fzimmermann89]

Ignore my previous question about fixture.
I prefer your current style.

For new functions and tests, please add type hints

[fzimmermann89]

I hope I did the merge correctly...