Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[MRG] FIX: sdml formulation and solvers (#162)
* FIX: make proposal for sdml formulation * MAINT clearer formulation to make the prior appear * MAINT call the prior prior * Use skggm instead of graphical lasso * Be more severe for the class separation * Put back verbose param * MAINT: make more explicit the fact that to use identity (i.e. an SPD matrix) as initialization * Add skggm as a requirement for SDML * Add skggm to required packages for travis * Also add cython as a dependency * FIX: install all except skggm and then skggm * Remove cython dependency * Install skggm only if we have at least python 3.6 * Should work if we want other versions superior to 3.6 * Fix bash >= which should be written -ge * Deal with tests when skggm is not installed and fix some PEP8 warnings * replace manual calls of algorithms with tuples_learners * Remove another call of SDML if skggm is not installed * FIX fix the test_error_message_tuple_size * FIX fix test_sdml_supervised * FIX: fix another sdml test * FIX quic call for python 2.7 * Fix quic import * Add Sigma0 initalization (both sigma zero and theta zero should be specified otherwise an error is returned * Deal with SDML making some tests fail * Remove epsilon that was unnecessary * FIX: use latest commit of skggm that fixes the non deterministic problem * MAINT: add message for SDML when not SPD * MAINT: add test for error message if skggm not installed * Try other syntax for installing the right commit of skggm * MAINT: make sklearn compat sdml test be run only if skggm is installed * Try another syntax for running travis * Better bash syntax * Fix tests by removing duplicates * FIX: fix for sdml by reducing balance parameter * FIX: update code to work with old version of numpy that does not have axis for unique * Remove the need for skggm * Update travis not to use skggm * Add a stable init for sklearn checks * FIX test_sdml_supervised * Revert "Update travis not to use skggm" This reverts commit 57b0567. * Add fallback on skggm * FIX: fix versions comparison and tests * MAINT: improve test of no warning * FIX: fix wrap pairs that was returning column y (we need line y), and fix the example for SDML to not raise another warning * FIX: force travis to do the right check * TST: add non SPD test that works with skggm's quic but not sklearn's graphical_lasso * Try again travis this time installing cython * Try to make travis work with build_essential * Try with installing liblapack * TST: fix tests for when skggm is not installed * TST: use better pytest skipif syntax * FIX: fix broken link in README.md * use rst syntax for link * use rst syntax for link * use rst syntax for link * MAINT: remove test_sdml that was remaining from drafts tests * TST: remove skipping SDML in test_cross_validation_manual_vs_scikit * FIX link also in getting started * Put back right indent * Remove unnecessary changes * Nitpick for concatenation and refactor HAS_SKGGM * ENH: Deal better with errors and skggm/scikit-learn * Better creation of prior * Simplification for init of sdml * Put skggm as optional * Specify skggm version * TST: make test about 1 feature arrays more readable * DOC: fix rst formatting * DOC: reformulated skggm optional dependency * TST: give an example for sdml_supervised with skggm where it indeed fails * TST: fix test that fails weirdly when executing the whole test file and not just the test * Revert "TST: fix test that fails weirdly when executing the whole test file and not just the test" This reverts commit 6f5666b. * Add coverage for all versions of python * Install pytest-cov for all versions
- Loading branch information