Skip to content

Optimization module improvements and new spline grids #143

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
valentjn merged 211 commits into from
Aug 5, 2019
Merged

Optimization module improvements and new spline grids #143

valentjn merged 211 commits into from
Aug 5, 2019

Conversation

@valentjn
Copy link
Member

@valentjn valentjn commented Jul 30, 2019 •
edited
Loading

This is for merging the changes made during writing my PhD thesis from the sgopt to the master branch.

Rough changelog:

  • base module:
    • New grid types:
      • NotAKnotBsplineBoundaryGrid
      • ModNotAKnotBsplineGrid
      • NaturalBsplineBoundaryGrid
      • FundamentalNotAKnotSplineBoundaryGrid
      • FundamentalSplineBoundaryGrid
      • WeaklyFundamentalNotAKnotSplineBoundaryGrid
      • ModWeaklyFundamentalNotAKnotSplineGrid
    • Corresponding basis types and evaluation operations
    • Hierarchisation for fundamental spline grids via breadth-first search
  • combigrid module:
    • Fix fallthrough warnings
  • datadriven module:
    • Add new grid types to LearnerSGDE and GridTypeParser
  • optimization module:
    • Fix various bugs in optimizers, move gradient members to base classes, add missing getters/setters
    • New framework for fuzzy extension principle (fuzzy logic) as an example application of SG optimization
    • New IterativeGridGenerator for fuzzy logic based on Ritter-Novak
    • Add convience bulk eval methods to ScalarFunction/VectorFunction and gradients
    • New ScaledScalarFunction function type and gradients
    • Add new grid types to optimization classes with respect to grid generation and hierarchization
    • Changes to test functions (names and equations)
  • pde module:
    • Fix dependency of test_OperationMoment on pde
  • Other:
    • Fix wrong path order in SConstruct
    • Fix Python 2.x iteritems() call in cpplint
    • Add script for fixing relative include paths

TODO:

  • add all new grid types to other modules if missing
  • add documentation if missing
  • add Doxygen documentation to fuzzy example
  • add tests if missing

Fixes #63.

valentjn added 30 commits August 1, 2016 09:25
@valentjn
Merge master into sgopt
d8911b6
@valentjn
Add Python 3.x support for pysgpp
d48a0e8 
Caveats:
* Disables Python tests.
* Searches Python 3.x include path with a rather dirty hack
* NumPy interface for DataVector/DataMatrix not tested
@valentjn
Merge master into sgopt
bae0b0b
@valentjn
Fix wrong naming of IsValidSequenceNumber
fa934da
@valentjn
Add Lagrange, not-a-knot, and natural bases
15f325e
@valentjn
Add support for spline degree up to 7
6b4d935
@valentjn
Add grids for new bases
7049109
@valentjn
Add OperationEval for new grids
2aab6db
@valentjn
Add derivatives for Lagrange bases
6d583b6 
* Add Lagrange*BasisDerivX for X = 1, 2, 3.
* Use coefficients in scientific notation instead of using
  very long quotients (slightly more readable code).
* Add new bases to pysgpp and jsgpp.
@valentjn
Add BFS hierarchisation operations for new bases
f50b15a
@valentjn
Remove hierarchisation for non-Lagrange bases
82342e9 
Have been too eager with my last commit.
@valentjn
Add mod. Lagrange not-a-knot splines + grid + op.s
74d4af5
@valentjn
Add EvalGradient op.s for new bases
5a67174
@valentjn
Add Hessian/part. deriv. op.s for new bases, tests
9c7bbf6
@valentjn
Remove hier. op.s for Lagrange bases
deaef63 
Theoretically, Lagrange bases are easier to be hierarchize.
Practically, things are not that easy as for each hierarhical
level, a linear system has still to be solved. The system should
be banded (d = 1) or at least sparse (d > 1); but for now, it's
too complicated to implement that. So, remove the Hierarchisation
operations from sgpp::base and add naive MultipleHierarchisation
operations in sgpp::optimization.
@valentjn
Fix pysgpp bug with SLagrangeNotAKnotSplineBase
5264b06
@valentjn
Add evaluation for even-index NAK B-splines
1e8cf7a
@valentjn
Allow setting of obj. fun. in optimizers
824ff92
@valentjn
Fix OpenMP bug in MultiStart (missing critical)
d00c598
@valentjn
Move gradient/Hessian to UnconstrainedOptimizer
2d4bd61
@valentjn
Fix bug in MultiStart's copy constructor
ba667c8
@valentjn
Propagate changed fcn. in MultiStart to optimizer
80f8cfe
@valentjn
Add first implementation of fuzzy functions
03eb6ca 
TODOs:
* Add to SWIG interfaces.
* Add tests.
* Add comments/documentation.
* Do a little more testing (other functions, higher
  dimensionalities, other grids, ...).
@valentjn
Propagate changed MultiStart gradients/Hessians
94be268
@valentjn
Fix bug in Hessian setter of Unconstr.Optimizer
d28a873
@valentjn
Add MC calculation of fuzzy interval norms/errors
19c324e
@valentjn
Make sure fuzzy ext. princ. outputs fuzzy interval
4a0b2eb
@valentjn
Add fuzzy classes to Python/Java SWIG interface
e4980cb
@valentjn
Compute errors in fuzzy example, better structure
fedd995
@valentjn
Use binary search in InterpolatedFuzzyInterval
04f479b
valentjn added 11 commits August 4, 2019 10:41
@valentjn
Disable -Wmisleading-indentation on mingw-w64
2cab714 
GCC sometimes fails with internal compiler error on mingw-w64
(e.g., "in gt_ggc_m_S, at ggc-page.c:1474" or "segmentation
fault"). Disabling -Wmisleading-indentation (from -Wall) seems to
fix the problem. Although GCC says when the warning is not
disabled that "pysgpp\pysgpp_wrap.cc:322190:0: note:
-Wmisleading-indentation is disabled from this point onwards,
since column-tracking was disabled due to the size of the
code/headers", it still crashes...
@valentjn
Rename testSLE -> test_SLE for consistency
63a37fd
@valentjn
Move unit tests of Printer and RNG to base
45fc35a
@valentjn
Use Printer for printing IterativeGridGenerator
cd79b31
@valentjn
Fix norm computation of FuzzyInterval
f61dee9
@valentjn
Add possibility to change binary search tolerance
f10ab50
@valentjn
Clarify docs of QuasiGaussianFuzzyNumber
0805f0a
@valentjn
Add parameter for number of integral samples
09e07bc
@valentjn
Add copy constructors for fuzzy intervals
5662b65
@valentjn
Fix quadrature tests
a28f8cb
@valentjn
Fix copy constructors of fuzzy extension princ.
58d2b58
MichaelReh
MichaelReh previously approved these changes Aug 5, 2019
View reviewed changes
Copy link
Member

@MichaelReh MichaelReh left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good to me.
However, this pull request is so large, that I could not review everything in detail.

@valentjn
Copy link
Member Author

valentjn commented Aug 5, 2019

@MichaelReh Thanks. Sorry that the PR is too big. The branch is a remnant of the SVN days at TUM, and has since been carried over to GitLab and then to GitHub (of course with some merging in between). I'll use smaller feature branches in the future.

@valentjn valentjn requested a review from MichaelReh August 5, 2019 14:08
@valentjn
Copy link
Member Author

valentjn commented Aug 5, 2019

This is now ready to merge. #157 is already contained in here and can be closed, if this is merged.

@valentjn valentjn merged commit 166aacf into master Aug 5, 2019
@valentjn valentjn deleted the sgopt branch August 5, 2019 16:34
@valentjn valentjn mentioned this pull request Aug 5, 2019
Closed
@leiterrl leiterrl added this to the v3.3.0 milestone Aug 6, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@freifrauvonbleifrei freifrauvonbleifrei freifrauvonbleifrei approved these changes

@MichaelReh MichaelReh Awaiting requested review from MichaelReh

Assignees

@valentjn valentjn

Labels

None yet

Milestone

v3.3.0

Development

Successfully merging this pull request may close these issues.

Choose inner optimization method in constrained optimization

5 participants

@valentjn @MichaelReh @freifrauvonbleifrei @leiterrl