Changelog¶
0.6.4 - Bugfixes¶
- Replaced usage of deprecated
scipy_mode
. Fixed #39 - Fixed
ValueError: invalid literal for int() with base 10
inpredict_qd
. Fixed #40 - Added input validators to raise human-readable error messages when the input is not correct. Fixes #37
- Fixed
AttributeError: module 'numpy' has no attribute 'object'.
. Fixes #38
0.6.3 - Bugfixes¶
- Fixed
ValueError
with recent versions ofSciPy
, due to usage of sparse arrays with object dtype. Fixes #31 - Fixed
IndexError
whenNaN
values are present in the dataframe. Fixes #28
0.6.2 - Warning filter¶
- Now filtering
UserWarning
infit_selector_model
even in the sklearn adapter. Fixes #20
0.6.1 - Minor changes¶
- Now filtering
UserWarning
infit_selector_model
. See #20 - New build system: now using
virtualenv
instead ofconda
innox
sessions. Fixes #23 - New project layout to avoid bug with
xunitparser
. Fixes #18
0.6.0 - sklearn api renames¶
-
Migrated CI/CD from Travis to Github Actions +
nox
. -
selector_skl
module renamedsklearn
andQDSSelector
renamedQDScreen
. Fixes #16
0.5.0 - First public working release¶
Initial release with:
- A main method
qd_screen
to get the (adjacency matrix of) the quasi-deterministic-forest, aQDForest
object with string representation of arcs (Fixes #8). - Possibility to
keep_stats
so as to analyse the (conditional) entropies in order to define a "good" threshold. - A method
<QDForest>.fit_selector_model(X)
to fit aQDSelectorModel
feature selection model able to select relevant features and to predict missing ones. Fixes #7 - Support for both pandas dataframes and numpy arrays as input. Fixes #2
- A Scikit-learn compliant feature selector
QDSSelector
, providing the exact same functionality as above but compliant with scikit-learnPipeline
s. Fixes #1
Non-functional: