Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[ENH] New transformation based pipeline classifiers (#1721)
* typo * fitted logic in panel case * boxcox refactored as example * linting * linting * added new tags * added checks for object dtype in Series types * fixing if/else logic path 2 and 3 * added nested_univ to df-list conversion * and back * changed fit-in-transform default to True * linting * linting * improved informativity of estimator tags error message * added more tags * fixed missing tag in boxcox refactor * random interval transformer * updated fit check with check_is_fitted * removed superfluous comments * remove return_pred_int errors * corrected data frame check * added inner types for panel trafos * output conversion for Series/Primitives * signature fixes * linting signatures * linting * linting signatures * stc transform saving * signatures update * typo * coerce to pandas * japanese vowels test * Fixing error on parameters for experiment * relaxing numpy series mtype * new rotation forest * remove useless atts * rotf contracting, train estimate and comments * doc fix * rotf example fix * stc comments p1 * include rotation forest * hc refactor p1 * hc1 test fix * stc rotf fix * cboss train probs * use transformed data in stc * hc1 test outline * config change * hc2.0 * rotf seed fix * st * st fix * stc and st tests and finishing touches * remove mstc dev * more st fixing * import code quality * rotf code quality * hc1 test code quality * hc2 tests and contracting * hc2 stc default * hc2 stc default (fix) * st rename and test fixes * hc2 train estimate fix * stc test fixes and codeowners * rotf fix * remove identical shapelets at end and transform n_jobs * rounding ST ig to prevent differences between OS * rotf test fix * st and stc docs * hc1 and hc2 test fixes * hc1 and hc2 examples fix * higher max shapelets for tests * more mv cases in tests * hc test parameters * hc test parameters 2 * hc test parameters 3 * hc test parameters 4 * wrong probas * hopefully done * dumb linalg packages ruin my week by causing differences between os * experiments update and a couple of getattr bugfixes * contrib experiments update * set classifier hc2 * base * enforce univariate in base class * remove unnecessary classifier tags * predict and predict_proba * tweaks to classifier base class * formatting 1 * formatting 3 * formatting 4 * formatting 6? * blank lines or no blank lines? * remove unnecessary argument to get_tag * negate tag correctly, remove unnecessary get_tag argument * correct tag negation * _predict _predict_proba _predict made abstract, _ predict_proba given a default implmentation n_classes_ added to base class * formatting 1 * formatting 2 * renamed fitted_trafos to transformers * deprecation warning category, bump version number * linting * linting * linting * linting * added defensive assert * added error message to defensive assertion * added transform-input tag in docstring * wrong tag used, now "univariate-only" * corrected reference in ForecastingPipeline * clarified mtype in docstring * linting * fixing transformer tests to correctly refer to create_test_instance * Revert "fixing transformer tests to correctly refer to create_test_instance" This reverts commit ad6877f. * transformer extension template * extension template docstrings * HC comments an experiments fixes * added link to transformer extension template in README * remove deprecated versions and more feature based updates * test config * feature based test tolerance * clearer mtype/scitype tracking logic * minor update to extension template * refactored input checks * bug in check_is function fixed * linting * linting * removing mtypes with unsupported checks * fixed typo in signature method * signature _fit should ignore y * added missing docstring in _series_as_panel/_convert * fixed docstring * corrected docstring * added valueerror message and capture for dim 1 np.ndarray in series/panel conversions * scitypes condition * linting * changed variable names in fit * bugfix in output conversion in transform * added comments suggested by Lovkush * remove new transform classifiers * signature and mp tests * signature example * reintroduce new transformers/transform classifiers * only mv summarycls is broken now * replaced fit-copy with fit-reference * Update extension_templates/transformer.py Co-authored-by: Lovkush <lovkush@gmail.com> * clarify special case * clarified * clarification on output type * fixed test reference * replaced transform(Z) reference by X reference in test_date * moved comments before lines * numpydoc compliance in extension template * clarified imports * added clarification on init * est2 fixed * removed fixed arg for transform testing * refactor summarizer and base class * linting * wrong place * output for transformer fix * had wrong X_inner_mtype * fixed test_raises_not_fitted_error test * changed fit-in-transform behaviour to "fit must always be called even if empty" * fresh prince * fresh prince 2 * i have to make this commit to swap branches and fix a bug * fix secondary error caused by changing fit-in-transform behaviour * testing changes, still some train estimate stuff to do * fresh prince test * summary classifier fix and docs * pre-pr bug fix * catch22 replace nans * catch22 single feature public * catch22 nan replacement change * tsfresh link Co-authored-by: Franz Király <f.kiraly@ucl.ac.uk> Co-authored-by: a-pasos-ruiz <56823538+a-pasos-ruiz@users.noreply.github.com> Co-authored-by: Alejandro Pasos Ruiz (CMP - Postgraduate Researcher) <fbu19zru@UEA.AC.UK> Co-authored-by: Tony Bagnall <ajb@uea.ac.uk> Co-authored-by: Lovkush <lovkush@gmail.com>
- Loading branch information