tice-ml 1.0.4

Model agnostic Time-Series Counterfactual Engine.

TiCE

TiCE is an extension of the DiCE <https://github.com/interpretml/DiCE>_ (Diverse Counterfactual Explanations) library, designed to provide counterfactual explanations with additional functionality and improvements. This library was developed as part of my master's thesis at TU Eindhoven while working at ASML.

This package is especially tailored for regression and time series applications, while maintaining compatibility with DiCE's original design principles.

Acknowledgement

This work builds upon the excellent DiCE <https://github.com/interpretml/DiCE>_ library created by the team at InterpretML.
I acknowledge and thank them for their contributions to the field of explainable AI.

The improvements in TiCE extend DiCE's capabilities while staying true to its core mission: generating actionable and diverse counterfactual explanations.

Improvements in TiCE

Compared to the original DiCE implementation, TiCE introduces the following enhancements:

• Support for regression tasks: Generate counterfactuals not only for classification but also for regression settings with a continious target variable.
• Time series compatibility: Adapted internal structures to handle sequential data, enabling counterfactual explanations for time-dependent models.
• Improved data interface: More flexible handling of continuous, categorical, and temporal features in heterogeneous datasets.
• Advanced Visualization: Dedicated visualization utilities tailored to time-series explanations are added. These visualization tools transform hundreds of numeric scores into digestible figures, making it far easier for domain experts and stakeholders to interpret model explanations and counterfactual suggestion

Installation

You can install TiCE directly from PyPI::

pip install tice

Usage

The usage of TiCE follows the same structure as DiCE with minor adjustments::

import TiCE

dice_exp = TiCE.Dice(model, data_interface) counterfactuals = dice_exp.generate_counterfactuals(query_instance, total_CFs=5, desired_range=[value_min, value_max])