Implement HOMER #2
Labels
Projects
Comments
ChristianSch
added a commit
that referenced
this issue
Jun 2, 2018
* Added support for sparse X and y. Corrected small typo: "ytring" -> "ystring" * Added tests for sparse X and y. * Added a return state to the fit() method to comlpy with the usual interface of scikit-learn. * Cleanup: deleted unused variables, corrected variable name case, ... * fixed dataset fetching and listing; closes #57 * hotfix: removed standard library packages from requirements.txt to prevent typosquatting and malicious code execution * Fix np.zeros in rakelo.py (#76) This commit fixes the use of np.zeros, formerly np.zeroes, to resolve Issue #76 Author: ljvmiranda921 Email: ljvmiranda@gmail.com * add citation info test if slack commit log works * Update README.md (#74) This commit updates the README.md for this library. New sections were added: - A short description of the project - Features - Dependencies - Installation - Basic Usage - Contributing - Cite Two badges were also added: - PyPI badge - License badge Next steps (can only be done by project owner): - Add travis-ci badge once a successfull travis-build is implemented (owner only). - Add documentation badge (owner only) Author: ljvmiranda921 E-mail: ljvmiranda@gmail.com * initial travis setup * fix tests * remove pyc in travis * name container properly, pass MEKA_CLASSPATH * add travis * add slack notifications * trying multi-env travis * fixed igraph package name * add test requirements * add osx test build * fix linux py3 build name * fix osx homebrew repo name * add one more osx homebrew repo * enforce p3 on osx in travis * fix python osx problems on travis * pip2 instead of pip * cask remove oclint * pip3 instead of pip * explicit python3 on osx for travis * correc liac-arff req * add dtype to np.zeros * add a per label binarizer for quality measures, closes #84 * fix BRkNN top label number selection syntax error * mod fit for sparse y columns ensure that sparse y columns of shapes like (800,1) (returned from some classifiers etc) are converted properly to shape (800,) -otherwise bugs are thrown by some scikit validation functions. * proper processing of output matrix structures ensure proper back and forth conversions of y values shaped like (800,1) and (800,) - to avoid errors thrown by some scikit validation functions. np.ravel does not properly process some matrices unless they are first cast to arrays. * make the code much more readable * some more variables renamed to a more informative name * import issparse, reformat * import numpy * prediction transposition in CC is no longer required * fix returning 1d label vector and testing for that * fix meka io bytes/strings and decode if needed, reformat * enforce updating to current docker image * fix travis for meka tests * add self-edge normalization option and fix test * disable weight normalization on unweighted graph * separate graph builders and label space clusterers, more tests written, some parameter sets for graphtool do not work atm * formatting changes * don't build a list if noone needs it * fix some circularity problems with typing * less extensive data testing for now, a lot of cases fail with certain generator params, due to one-classiness of partitions * fix stochastic block modelling based on graphtool * fix and standardize clustering output alongside with a proper integration test in label space partitioning classifier * remove osx travis for now, not working anyway * adjust partitions to new test data * adjust labelset sizes to new test data * make sure correct version is pulled * change the default rakeld/rakelo behaviour to include labelpowerset, a voting classifier is added to allow overlapping classification in rakel style with any clusterer and clasifier, a rakel's clustering logic is moved to a random clusterer * fix CV base test * travis python2 should work correctly now, with new devdocker * introduce a working test case instead of randomly crashing generators * add absoulte imports to fix igraph import in p2 * fix tests and add set_params support for clusters so that CV works * some flaky setup line for travis * fix random label space clusterer test with overlaps to pass * temporary workaround for dense matrices * workaround output in matrix shape as well * update documentation * documentation and naming corrections * fix rename of cluster_*
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Implement infrastructure for hierarchical classifiers and HOMER
The text was updated successfully, but these errors were encountered: