Releases: haesleinhuepf/apoc
Releases · haesleinhuepf/apoc
0.12.0
New features
- The
ObjectSelector, based on theObjectClassifierfor selecting objects of a given class. It produces an instance segmentation label image. - The
ObjectMergerhas new options for features:centroid_distance,mean_intensity_difference,standard_deviation_intensity_difference,area_difference,mean_max_distance_to_centroid_ratio_difference
Potentially backwards breaking changes
- Internal refactoring in
ObjectClassifierandTableRowClassifiermay have caused backwards-compatibility breaking changes. This is not intentional. In case anyone experiences trouble, please create a github-issue and we'll fix it.
0.11.0
0.10.0
Bugfixes
- Fix File not found error when training on folders (thanks to @jo-mueller for reporting)
- Touching-neighbor-count may be weighted differently when training because of upstream changes in pyclesperanto-prototype 0.19.0
Full Changelog: 0.9.0...0.10.0
Contributors
jo-mueller
Assets 2
0.9.0
New features
- enabled neighbor statistics for objec classification
Backwards compatibililty breaking changes
- apoc now depends on pyclesperanto_prototype >= 0.18.3
0.8.1
What's Changed
- Print classifier information when calling
str(classifier)or> classifieron the ipython console by @jo-mueller in #31 - Feature importance by @haesleinhuepf in #30
Bug fixes
- Unknown features cause an exception now in the ObjectClassifier
- When older .cl files were read, this caused errors while loading because some fields were not filled with default values. Old .cl files should now work again,
- Added missing dependency: pandas
New Contributors
- @jo-mueller made their first contribution in #31
Full Changelog: 0.8.0...0.8.1
Contributors
haesleinhuepf and jo-mueller
Assets 2
0.8.0
Changed
- The default value for
num_ensembleshas been increased to 100 and is now in line with scikit-learn defaults. This higher number of trees makes results more reproducible.
0.7.0
New features
TableRowClassifier.train()can consume table withNaNelements, which will be ignored while trainingTableRowClassifier.train()can consumeground_truthlist entries==0which will be ignored while training- Added a new predefined feature set
v070
Backwards compatibility breaking changes
TableRowClassifier.predict()has no parameterreturn_numpyanymore and returns a numpy array.- Input features and results of
TableRowClassifiertrain()andpredict()are now always 1-dimensional. 2D arrays with shape(1, n)are no longer allowed.
Thanks to @kevinyamauchi for feedback and support!
Contributors
kevinyamauchi
Assets 2
0.6.8
New features:
- Measurements and
APOC_ObjectClassification_CLUSTER_IDare stored inclassifier._dataafter object classification
0.6.7
New features
TableRowClassifierfor object classification beased on custom measurement tables (Thanks to @kevinyamauchi for implementing this)
Contributors
kevinyamauchi
Assets 2
0.6.6
Bug fix
- Type error in error message when a wrong type of classifier is loaded