v1.1.3

Hotfixes.

1.1.2

Hotfixes.

1.1.1

• Addressed build issues.

v1.1.0

teex v1.1.0 (#5 )

(Package)

• Two new datasets in saliencyMap module: CUB-200-2011 and Oxford-IIIT Pet.
• New utility methods in saliencyMap datasets: delete_data, get_class_observations
• Nice bug fixes

(Build & Docs)

• CI improvements: code coverage, automatic semantic release
• Documentation: new documentation theme and .ipynb support

(Tests)

• Increased coverage significantly

1.0.4

Minor bugs squashed.

1.0.3

Changes

Feature Importance

• New function lime_to_feature_importance in featureImportance.data that converts from a lime.explanation.Explanation object to a np.array feature importance vector.
• Scaler function for feature importance is now a public method in the data module of the feature importance sub package.
• Automatic scaling in featureImportance.eval.feature_importance_scores now performs on a per-column basis.

Decision Rule

• clean_binary_statement, a method for parsing Statement edge cases is now a public method in the data module.

...and other minor bug fixes.

1.0.1

Feature Importance bug fixes:

• FI explanations now scale to (-1, 1) by feature
• FI explanations do not raise KeyError when not appearing in the expression's derivative.

teex v1.0.0

First major release 🚀 Major API changes for usability from previous version.

• Subpackage system overhaul: subpackages for each explanation type containing data and eval modules.
• Synthetic dataset system overhaul: behaviour matched to the real datasets; they are now implemented as sliceable objects.

Moreover...

• Word importance scores implemented
• Added unittest suites with tests for all subpackages: data and metrics
• New functionalities to the DecisionRule class:
  • rulefit to decision rule
  • string to decision rule parser
• Support changed to Python >= 3.6
• Statement class usage change
• Minor bug fixes

❤️

Alpha 1

Fixed package building issues