Skip to content
Commits on May 29, 2008
  1. @hanke

    Merge branch 'master' into debian

    hanke committed May 29, 2008
    * master:
      BF: Have to enable libsvm to get full source tarball.
  2. @hanke
  3. @hanke

    Merge branch 'master' into debian

    hanke committed May 29, 2008
    * master:
      BF: Fix debsrc makefile target.
      BF: Let make orig-src work on non-Debian branches.
  4. @hanke

    BF: Fix debsrc makefile target.

    hanke committed May 29, 2008
  5. @hanke
  6. @hanke

    Merge branch 'master' into debian

    hanke committed May 29, 2008
    * master:
      NF: Add Makefile target to create Debian source package.
  7. @hanke
  8. @hanke
  9. @hanke
  10. @hanke

    Release debian version 0.2.0-1.

    hanke committed May 29, 2008
  11. @hanke

    Merge branch 'master' into debian

    hanke committed May 29, 2008
    * master: (24 commits)
      BF: Missed release date by one day.
      RF+PL: wr1996 dataset generation moved under data_generators
      automagically uptodated tables of state variables
      BF: small fix in tools/fillin_states
      DOC: few pieces here and there
      RF: proper assignment of default values to 'helper' SVM defintions
      RF:  testapiref depend on apidoc, more phonies
      RF: Warehouse became independent of classifiers class
      NF: "make testapiref" checks present references in documentation
      RF: SVMs refactoring
      RF: lower needed % correct in the test since we hit the bottom some times
      RF: we should allow working PyMVPA with no SVM implementations
      disabled 1 test since it is invalid if data is simple
      RF: both libsvm and sg got the same top class name SVM for consistency
      BF: SVRLight seems to be absent from some shogun releases, make it conditional
      BF: Fixing some of the broken links to api documentation
      BF: Fix broken API link.
      BF: Mention GPR in changelog, minor change to Makefile
      RF: tuned up datasets for testing/training and used them through test_clf
      NF: added constant and linear kernels
      ...
  12. @hanke
  13. @yarikoptic

    RF+PL: wr1996 dataset generation moved under data_generators

    yarikoptic committed May 28, 2008
    misc pylint-friendly fixups in data_generators.py
  14. @yarikoptic

    Merge branch 'yoh/master' of ssh://alioth.debian.org/git/pkg-exppsy/p…

    yarikoptic committed May 28, 2008
    …ymvpa into yoh/master
    
    * 'yoh/master' of ssh://alioth.debian.org/git/pkg-exppsy/pymvpa:
      automagically uptodated tables of state variables
      BF: small fix in tools/fillin_states
  15. @yarikoptic
  16. @yarikoptic

    BF: small fix in tools/fillin_states

    yarikoptic committed May 28, 2008
    TODO: Make it do the same for parameters?
  17. @yarikoptic

    Merge commit 'alioth/mh/master' into yoh/master

    yarikoptic committed May 28, 2008
    * commit 'alioth/mh/master':
      BF: Fix broken API link.
      BF: Mention GPR in changelog, minor change to Makefile
  18. @yarikoptic
  19. @yarikoptic
  20. @yarikoptic
  21. @yarikoptic

    RF: Warehouse became independent of classifiers class

    yarikoptic committed May 28, 2008
    so we could use it for something else later on
Commits on May 28, 2008
  1. @yarikoptic
  2. @yarikoptic

    RF: SVMs refactoring

    yarikoptic committed May 28, 2008
    * now mvpa.clfs.svm is the one which creates convenience classes
      (e.g. LinearCSVMC) based on the default backend (libsvm is the
      default if found). Thus there is mvpa.clfs.svm.SVM, as well as
      mvpa.clfs.svm.LinearCSVM
    
    * It is preferable to use simply SVM for the classifier name and
      parametrize it appropriately (ie kernel_type, svm_impl, etc)
    
    * now to get access to backend SVM implementations, use
      mvpa.clfs.<backend>.SVM
    
    * sensitivities callbacks for linear SVM are assigned within _KERNEL
      and spit out by _SVMBase itself, so we don't duplicate things
    
    TODO: docstrings
  3. @yarikoptic
  4. @yarikoptic

    RF: we should allow working PyMVPA with no SVM implementations

    yarikoptic committed May 28, 2008
    warning instead of raise
  5. @yarikoptic

    disabled 1 test since it is invalid if data is simple

    yarikoptic committed May 28, 2008
    and test barely can spot out any problem
  6. @yarikoptic
  7. @yarikoptic
  8. @yarikoptic
  9. @hanke

    BF: Fix broken API link.

    hanke committed May 28, 2008
  10. @hanke
  11. @yarikoptic
  12. @yarikoptic

    Merge commit 'alioth/eo/master' into yoh/master

    yarikoptic committed May 28, 2008
    * commit 'alioth/eo/master':
      NF: added constant and linear kernels
      RF: maximization of log marginal likelihood has been splitted in 2
      NF: KernelSquaredExponential now performs Automatic Relevance Determination.
      PL: removing trailing whitespaces.
      BF: changed ModelSelector accordingly to gpr.py
  13. @emanuele

    NF: added constant and linear kernels

    emanuele committed May 28, 2008
    Added two new kernels (constant and linear). A simple
    test is in model_selector.__main__ .
    Improved some docstrings.
  14. @emanuele

    RF: maximization of log marginal likelihood has been splitted in 2

    emanuele committed May 28, 2008
    methods
    
    In ModelSelector, in order to maximize the log marginal likelihood now
    you need to perform 2 steps: 1) max_log_marginal_likelihood() to set
    up the minization problem, 2) solve(), to do the actual minimization
    step via OpenOpt. This is useful since step 1 returns an instance of
    of the problem to be minimized, which can be tweaked by the user,
    e.g., adding custom constraints. Then step 2 does the job accordingly.
    
    In __main__ added some test code to investigate Automatic Relevance
    Determination through maximization of log marginal likelihood.
Something went wrong with that request. Please try again.