Fixes:
- Add missing dependency on
packaging
(#247). Thank you @sim-san for reporting and fixing! - Fixed an issue causing routed parameters (
fit__class_weight
) to be overridden by non-routed parameters, including default values (class_weight=None
) (#244). Thank you @8549 for reporting the issue. - Fix handling of
class_weight="balanced"
(#245)
Features:
- Support routed parameters to callbacks (#233)
Changes:
- Official support for extra keyword arguments (
**kwargs
) infit()
,predict()
andpredict_proba()
that are routed directly to Keras (#238) - Remove hard dependency on
tensorflow
, allowing users to choosetensorflow-cpu
or other distributions in environments where they can't controlpip
arguments (#237) - Remove hard dependency on
tensorflow
, allowing users to choosetensorflow-cpu
or other distributions (#237)
Fixes:
- Fixes a bug in inverse-transforms of predictions for multiclass labels that are not one-hot encoded (#229). Credit to RKCZ on StackOverflow #67019157 for reporting the issue.
Thank you to all of the collaborators that found bugs and submitted PRs to SciKeras!
v0.3.0 is a minor release that consists mainly of bug fixes, although we have made huge improvements under the hood as well.
Features:
- Implement batch_size=-1 (#194)
- Lots of documentation improvements (#200, #197, #178, #176, #174)
Fixes:
- Fix a bug in meta parameter collection (#171)
- Allow
epochs
to be passed as a keyword argument to fit (#154) - Fix the signature of
BaseWrapper.scorer
(#169) - Fix inverse transforms for one-hot encoded targets (#189)
Contributors to this release:
- @stsievert
- @data-hound
Thank you to @stsievert for your continued support and contributions!
Release notes:
- Support autoencoders and more general use cases via BaseWrapper (#123)
- Fix slowdown caused by sample_weight processing (reported in DaskML#764, fixed in #123)
- Documentation improvements (#134, #135, #145 and #138)
- Fix the
initialize
method in KerasClassifier (#140)
- Move data transformations to a Scikit-Learn Transformer based interface (#88)
- Add Keras parameters to BaseWrapper.init (loss, optimizer, etc) (#47, #55)
- Remove needless checks/array creation (#63, #59)
- Make pre/post processing functions public (#42)
- Some stability around
BaseWrapper.__call__
(#35) - Cleanup around loss names (#38, #35)
- Parameter routing (#67)
- Rename build_fn to model with deprecation cycle (#98)
- Add ability for SciKeras to compile models (#66)
class_weights
parameter (#52, #103)classes
param forpartial_fit
(#69, #104)- Provide an epochs parameter (#51, #114)
- Updated docs, now hosted on RTD (#58, #73)
- Checks to make sure models are correctly compiled (#86, #100, #88)
- Add support for partial fitting and warm starts (#17, thank you @stsievert!).
- Add support for random states via the
random_state
parameter (#27). - Scikit-Learn SLEP10 compliance (#26).
- Fix unnecessary data reshaping warnings (#23).
- Versioning fix.
- Rename repo.
- Derive BaseWrapper from BaseEstimator.
- Python 3.8 support.
- Deprecate Python 3.5 (#11).
- Fix prebuilt model bug (#5).
- Clean up serialization (#7).
- Implement scikit-learn estimator tests (#10).
- Offload output type detection for classification to
sklearn.utils.multiclass.type_of_target
. - Add documentation.
- Some file cleanup.
- First release on PyPI.