-
Notifications
You must be signed in to change notification settings - Fork 0
/
ChangeLog
95 lines (84 loc) · 3.91 KB
/
ChangeLog
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
v1.8:
- rewrote NonnegativeMatrixFactors to use a truncated SVD on a single matrix.
- several examples now may parse either a single matrix or two matrix factors.
- several examples now allow for solving split feasibility problems, i.e. to
have additional constraints besides the minimum norm solution such as Basso
and RangeProjector.
- MatrixFactorizer has an additional OpenMP parallelization.
- several important fixes to MPI parallelization implementation in
MatrixFactorizer.
- smaller convenience changes to MatrixToPNG.
v1.7:
- Database functionality is now completely extracted from GeneralMinimizer.
- "untemplated" FunctionalMinimizer slightly modified and added
FunctionalMinimizerFactory such that values controling functional
minimization can be set globally.
- added regression tests for written sqlite file.
- Added NonnegativeMatrixFactors for generating good starting values for the
nonnegative matrix factorization.
- fixed regression tests on nonnegative matrix factorization.
v1.6:
- enhanced Database functionality and abstraction to allow for sensible data
accumulation in multiple loops, e.g. MatrixFactorizer
- MatrixFactorizer can parse initial factors from file
- inner minimization thresholds can be set from command-line
- converted MatrixFactorization into an SFP problem with AuxiliaryConstraints
- all example programs admonish faulty command-line parameters in an
comprehensible way
- all program show version and git build on first line
- some fixes and extensions to doxygen documentation
v1.5:
- CT example may store noisy sinogram.
- added "rotate" option to MatrixToPNG.
- changes to Matrix Factorization:
- now done as Split Feasibility Problem.
- fixed MPI part.
- uses StoppingCriteria class as well.
- removed stuff that was inconsistent with general Basso base.
- extracted scaling ambiguity remover into own class.
v1.4:
- new example Gravity
- new "example" MatrixToPNG (tool for ComputerTomography actually)
- stopping criteria can be specified on command-line with boolean ops
- ComputerTomography can use true solution and add noise (inverse crime check)
- discrepancy parameter can now be used with any Minimizer, not just RESESOP
- FIX: Command-line options sensibility check
- FIX: install target was broken
- FIX: user-specified diff for regression tests possible, e.g. ndiff
v1.3:
- added ComputerTomography example.
- added RangeProjector example.
- added NoiseAdder example.
- linear mapping factory can now parse matrix file directly.
v1.2:
- several fixes to stepwidth used in regularized l1 norm, to soft thresholding,
and to MatrixFactorizer.
- moved all examples away from src folder into extra examples folder. Each has
now a base library and the main() inside a slim executable along with an
Options folder.
- initial doxygen documentation.
v1.1:
- orthogonalizable search directions.
- adjoint of a linear mapping is now more cleverly stored (i.e. as internal
transpose flag and not as separate copy of the matrix).
- database is now split into parameters and data with views to recreate the
old access style.
- all examples now understand XML-style configuration files.
- added first version of MatrixFactorizer example.
v0.4:
- Landweber is working with regularized l1 norm.
- refactored to hide visibility of Eigen interface to some extent.
- added regularized SESOP with two search directions and good starting values.
- all vector and matrix operations are now fully counted to know about the
computational complexity.
v0.3:
- added inexact line search using NLopt and based on Wolfe conditions.
- tried various stuff such as different order of search directions used,
checking on angles between search directions.
v0.2:
- refactored subspace method to be in line with Landweber and fixes.
- added initial regularized l1 norm with duality mappings, i.e. soft
thresholding operator.
- added factory for the minimizer.
v0.1:
- initial version with Landweber and subspace methods.