hermai 0.1.8

A hermeneutic approach to explainable AI (XAI) with contextual and narrative explanations.

Hermai: A Hermeneutic Explainable AI Library

Hermai is a Python library for explainable AI (XAI) that focuses on providing contextual, narrative, and actionable explanations for machine learning models. Inspired by hermeneutics, it aims to create a dialogue between the user and the model's decision-making process.

Key Features

• Contextual Perturbations: Generates realistic "what-if" scenarios that respect the correlations in your data.
• Local Explanations: Explains individual predictions by identifying key feature contributions and generating counterfactuals.
• General Explanations: Approximates the overall behavior of a complex model with a simple, interpretable surrogate model (like a decision tree).
• Narrative Summaries: Presents explanations in human-readable text.

Installation

You can install Hermai directly from GitHub:

pip install hermai

Quick Usage

Here's a quick example of how to get local and general explanations for a RandomForest model trained on the Titanic dataset.

import pandas as pd
import seaborn as sns
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from hermai.perturbations.tabular import TabularPerturbationGenerator
from hermai.explainers import LocalExplainer, GeneralExplainer

# Load and prepare data (assuming X_train, black_box_model are ready)
# ...

# --- Local Explanation ---
generator = TabularPerturbationGenerator(categorical_features=['pclass', 'sex', 'embarked'])
generator.fit(X_train)
local_explainer = LocalExplainer(black_box_model, generator)
instance = X_test.iloc[0]
local_explanation = local_explainer.explain(instance)
print(local_explanation.narrative())
local_explanation.plot()


# --- General Model Explanation ---
general_explainer = GeneralExplainer(black_box_model)
general_explanation = general_explainer.explain(X_train)
general_explanation.plot_feature_importance()
general_explanation.plot_surrogate_tree()

License

This project is licensed under the MIT License - see the LICENSE file for details.