mini-causal 0.4.21

A machine learning decision intelligence framework that helps practitioners understand how feature changes influence model behavior, performance and business decisions

Mini-Causal

Mini-Causal is machine learning decision intelligence framework that helps practitioners understand how feature changes influence model behavior, performance, and business decisions.

Key features

• Causality tools: compare models using causal metrics and tests.
• Counterfactuals: generate and evaluate counterfactual explanations.
• Partial counterfactuals: run targeted counterfactual analyses on subsets of features.
• Batteries of examples: runnable Jupyter notebooks demonstrating common workflows.

Getting Started with mini-causal in a notebook

• Create and activate a virtual environment (PowerShell example):

python -m venv .venv
.\.venv\Scripts\Activate.ps1

• Install the package:

pip install mini-causal

**Quick after you loaded the data with X and y sets:

import pandas as pd
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from mini_causal.partial_counter import PartialCounterfactualClassifier

X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2,random_state=42)

model=LogisticRegression(solver="liblinear")
model.fit(X_train,y_train)

counter_classifier=PartialCounterfactualClassifier(model=model,X=X_train,y=y_train)

## get the individual causal effects:
ind_causal_effect_age=counter_classifier.individual_causal_effects(X=X_train,feature="age")

##accuracy effect by the exclusion of the feature
age_acc_effect=counter_classifier.causal_effect_metric(X=X_train,
                                                       y=y_train,
                                                       metric="accuracy",
                                                       feature="age")

## recall effect
age_recall_effect=counter_classifier.causal_effect_metric(X=X_train,
                                                          y=y_train,
                                                          metric="recall",
                                                           feature="age")

##log-loss effect
age_log_loss_eff=counter_classifier.causal_effect_probs_metric(X=X_train,
                                                               y=y_train,
                                                               metric="log-loss",
                                                               feature="age")

## hamming-loss effect
age_hamming_loss_eff=counter_classifier.causal_effect_loss_metricX_train,y_train,metric="hamming-loss",feature="age")


## true negative effect
age_tn_effect=counter_classifier.causal_effect_confusion(X_train,y_train,metric="tn",
                                                                feature="age")

##true positive effect
age_tp_effect=counter_classifier.causal_effect_confusion(X_train,y_train,metric="tp",
                                                                feature="age")

##summarized tabular dataframe
tab_metrics_df=counter_classifier.causal_effect_metric_tab(X_train,
                                                           y_train,
                                                           feature="age"
)

Examples

This folder contains runnable Jupyter notebooks demonstrating mini_causal features.

Notebooks included

• — demonstrates the library in a regression workflow.
• — demonstrates the library in a classification workflow.

Citing mini-causal

If you use mini-causal in your research, please cite it:

@software{mini-causal,
  title = {mini-causal: A Decision Intelligence Framework for machine learning},
  author = {Masemene Matlakana Benny},
  year = {2026},
  url = {https://github.com/MasemeneMatlakanaBenny/Mini-Causal},
  doi = {10.5281/zenodo.19646175},
  license = {MIT},
}

See CITATION.cff for the full citation metadata.

Contributing

See CONTRIBUTING.md for guidelines on reporting issues, opening pull requests, code style, and testing.

License

This project is released under the terms of the MIT License.