shapash

What's Changed

Added

• Feature/contribution plot improvement by @guillaume-vignal in https://github.com/MAIF/shapash/pull/553
• Feature/shapash report improvement by @guillaume-vignal in https://github.com/MAIF/shapash/pull/564

Fixed

• Fix color style. by @MLecardonnel in https://github.com/MAIF/shapash/pull/561
• fix documentation generation bug due to numpy 2.0 by @guillaume-vignal in https://github.com/MAIF/shapash/pull/566
• Fix interaction plot bug on labels by @guillaume-vignal in https://github.com/MAIF/shapash/pull/563

Full Changelog: https://github.com/MAIF/shapash/compare/v2.5.1...v2.6.0

What's Changed

• Temporary Fix for NumPy 2.0 Incompatibility by @guillaume-vignal in https://github.com/MAIF/shapash/pull/559

Full Changelog: https://github.com/MAIF/shapash/compare/v2.5.0...v2.5.1

What's Changed

Breaking changes

• Dropped support for 3.8 in @MLecardonnel in https://github.com/MAIF/shapash/pull/538
• Support only Scikit learn>=1.4.0 by @MLecardonnel in https://github.com/MAIF/shapash/pull/543
• Support only pandas>=2.1.0 by @guillaume-vignal in https://github.com/MAIF/shapash/pull/551
• Support only shap>=0.45.0 by @guillaume-vignal in https://github.com/MAIF/shapash/pull/552

Added

• Feature python 3.12 support by @MLecardonnel in https://github.com/MAIF/shapash/pull/538
• Optimization: compute predictions and probabilities directly in the compile step by @guillaume-vignal in https://github.com/MAIF/shapash/pull/535, https://github.com/MAIF/shapash/pull/544

Fixed

• Fixed report demo by @MLecardonnel in https://github.com/MAIF/shapash/pull/540, https://github.com/MAIF/shapash/pull/541, https://github.com/MAIF/shapash/pull/546, https://github.com/MAIF/shapash/pull/547, https://github.com/MAIF/shapash/pull/548, https://github.com/MAIF/shapash/pull/549

Full Changelog: https://github.com/MAIF/shapash/compare/v2.4.3...v2.5.0

What's Changed

• remove code for category_encoder<=2.2.2 by @guillaume-vignal in https://github.com/MAIF/shapash/pull/530
• Hotfix shap 0.45.0 by @guillaume-vignal in https://github.com/MAIF/shapash/pull/534
• last release for: python 3.8, shap<0.45.0, scikit-learn<1.4

Full Changelog: https://github.com/MAIF/shapash/compare/v2.4.2...v2.4.3

What's Changed

• Feature/code quality by @guerinclement in https://github.com/MAIF/shapash/pull/521
• Bump dash from 1.9.1 to 2.15.0 by @dependabot in https://github.com/MAIF/shapash/pull/526
• Feature/lint by @guerinclement in https://github.com/MAIF/shapash/pull/522

New Contributors

• @guerinclement made their first contribution in https://github.com/MAIF/shapash/pull/521

Full Changelog: https://github.com/MAIF/shapash/compare/v2.4.1...v2.4.2

Fix #514 BUG: with version 2.4.0 TreeExplainer is never used

Features:

• Support for Python 3.11 #512
• Be able to use Shapash to compute Shapeley values through Shap for any model supported by Shap #506

Breaking change:

• Removes ACV from shapash and fixes dependencies #482

Fixes:

• #475
• #503
• #497

Hotfix for handle missing data for categorical columns #490

Fix this issues : https://github.com/MAIF/shapash/issues/489 https://github.com/MAIF/shapash/issues/478 https://github.com/MAIF/shapash/issues/472

Update extra require for report pip install shapash[report] #494

Unit tests of the webapp's callbacks have been created.
In addition, a first refactoring of the Webapp is done to simplify maintenance and future developments.
#463

Some fix:
#455 References and modifications of self in smartapp create unwanted behavior
https://github.com/MAIF/shapash/issues/483 Bug in Webapp for filter

Remove numpy and pandas version limits.
Shapash has shap and xgboost dependencies that have been released. We no longer need to set numpy and xgboost versions.

Fix #451, #457

Hotfix for shapash and shap incompatibility with numpy versions 1.24.0 and above. We limit numpy to a version under 1.24.0.

PR #459 resolves incompatibility between shapash code and numpy.
But Issue #457 remains because of shap incompatibility with numpy.
PR #460 limits numpy versions under 1.24.0.

Hotfix for dash incompatibility with Flask versions 2.3.0 and above. We limit Flask to a version under 2.3.0.

Hotfix for pandas 2.0.0, however, we limit pandas for the moment because some tests do not pass with xgboost.

The main bug on the app with pandas 2.0.0 is fixed
#451 fix pandas version until xgboost is fixed and fix other code for pandas

Other fixes:
#440 Remove data from package
#441 scatter plot prediction when not enough data

We have a new tutorial and a new dataset:
#443 Feature/tutorial accident

These 2 new features are designed to understand its model just by browsing the Webapp and have all the necessary information

• With additional dataset columns to have other contextual information than the features of the model
• With an Identity card to better see characteristics of a single sample

✨ Features
#422 Feature/webapp visuals
#424 Feature/id card
#425 Feature/additional data
#426 Target and error columns in dataset

⬆️ Upgrade dependencies and stop support for Python 3.7
#418 maximum version for category_encoders and bump version
#414 maximum version for sklearn
#421 Feature/fix sklearn ce dropping Python 3.7

🐛 Bug fixes
#428 Selecting an index via the Index input box for integers on Windows fails
#429 Selecting an invalid index via the Index input box logs an error
#430 Local explaination plot fails on positive/negative contributions display for specific cases
#433 SmartExplainer method init_app fails when no y_pred
#431 Errors are not managed when manipulating filters on the Shapash Webapp
#432, #434 Update python version for docs

Fix maximum version for category_encoders waiting for change to adapt new version 2.6.x:
#418 maximum version for category_encoders

This patch release fixes several bugs on webapp:
#403 Webapp : when zooming, labels keeps the short format with "..."
#405 Minor bug on Webapp, When filter and zoom on contribution_plot for global population
#406 Webapp: A small bug with groups of variables and selecting a point
#415 Webapp: bug when click on a single sample, it removes the sub-selection of the feature importance

And fix maximum version for sklearn waiting for change to adapt new version 1.2.x:
#414 maximum version for sklearn

These 2 new features are designed to select samples in the Webapp

• With a new tab "Dataset Filter" to filter more easily with the characteristics of the features
• With a graph that represents the "True values vs Predicted values"

✨ Features
#389 Webapp: Improve the top menu for class selection
#388 Create to tab which contains prediction picking graph and connexion with other graph
https://github.com/MAIF/shapash/issues/387 add responsive titles, subtitles, axis titles and axis labels to each graph
https://github.com/MAIF/shapash/issues/386 Add explanation button and popup
https://github.com/MAIF/shapash/issues/385 Adapt the labels of graphs according to their size
https://github.com/MAIF/shapash/issues/384 Add tab that contains dataset filters
https://github.com/MAIF/shapash/issues/378 Adding a plot to the webapp and interactivity with other plots
https://github.com/MAIF/shapash/issues/377 Add of a prediction error distribution graph

✨ Features
New feature #376
Clustering of the correlation matrix in order to visualize correlations between variables easily.

New support in Python version
⬆️ Support to Python 3.10 #297
⬇️ Stop to support Python 3.6 #297

Upgrade dependencies
⬆️ scikit-learn>=0.24.0 #297
⬆️ acv-exp>=1.2.0 #297
⬆️ category_encoders>=2.2.2 #372

✨ Features
New feature #364
Pairwise comparison of Consistency : How are differences in contributions distributed across features ?