Skip to content

Latest commit

 

History

History
72 lines (53 loc) · 3.69 KB

README.md

File metadata and controls

72 lines (53 loc) · 3.69 KB

Experiments for the article titled: Factual and counterfactual explanations in fuzzy classification trees

Installation

Datasets

The datasets used in the experiment are the following ones:

DATASETS PARAMETER
Beer beer
Breast breast
Compas compas
Heloc heloc
Pima pima

Execution

Illustrative example

For the illustrative example of Section VII refer to the Jupyter Notebook illustrative_example.ipynb

Experiment 1: Lambda and Beta Study

For the experiment carried out in Section VIII.D.1 and Section VIII.D.2, the experiments are run by the file lambda_beta_study.py as follows:

$ python lambda_beta_study.py db -q?

Where db can take any of the values of the datasets previously explained and -q is an optional parameter to make a small sample of the database for quick results. The flag -q has not been used to generate the tables of the article and it is included for debugging purposes only.

This program will generate a file lambda_beta_study/db.csv that has the structure:

kwargs length q_multiple nr_fact
lambda and beta arguments length of the factual more than a single rule non robust factual percentage

In order to generate the plots for Table II and Fig. 2, refer to the Jupyter notebook generate_multiplots.ipynb.

Experiment 2: Factual Study

For the experiment carried out in Section VIII.D.3, the experiments are run by the file factual_study.py as follows:

$ python factual_study.py db -q?

Where db can take any of the values of the datasets previously explained and -q is an optional parameter to make a small sample of the database for quick results. The flag -q has not been used to generate the tables of the article and it is included for debugging purposes only.

This program will generate a file factual_study/db.csv that has the structure:

fact_name length q_multiple nr_fact
Name of the factual method length of the factual more than a single rule non robust factual percentage

In this case, each db.csv represents a different column of Tables III and IV, where if nr_fact > 0 the column is in italics.

Experiment 3: Counterfactual Study

For the experiment carried out in Section VIII.E, the experiments are run by the file cf_study.py as follows:

$ python cf_study.py db -q?

Where db can take any of the values of the datasets previously explained and -q is an optional parameter to make a small sample of the database for quick results. The flag -q has not been used to generate the tables of the article and it is included for debugging purposes only.

This program will generate a file cf_study/db.csv that has the structure:

cf_name n_changes n_rules
Name of the factual method Number of changes made to the instance Number of rules tested by the algorithm

With a last row that represents the number of counterfactuals.

In this case, each db.csv represents a different column of Table V, where the last row of the file represents the NumCF

Experiment 4: Study against existing proposals

TBD