influential-analysis 0.1.2

A library for finding influential instances in ML models

Influential Analysis

📌 Overview

Influential Analysis is a Python library for identifying and analyzing influential instances in machine learning models. It helps assess model accuracy and fairness by detecting data points that significantly impact predictions. This is especially useful for bias detection and fairness auditing.

🔥 Features

• Identify influential instances that impact model fairness & accuracy
• Supports classification models (e.g., Decision Trees, Random Forests, etc.)
• Uses Fairlearn for fairness assessment
• Provides visualization of influential data points

📦 Installation

Install the package using:

pip install influential-analysis

🚀 Usage

1. Import the Library

!pip install influential-analysis==0.1.1
from influential_analysis.influential_analysis import InfluentialInstanceAnalyzer

2. Train Your Model

from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
import pandas as pd

# Load Data
df = pd.read_csv("your_dataset.csv")
X = df.drop(columns=["target_column"])
y = df["target_column"]

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

# Train Model
model = DecisionTreeClassifier()
model.fit(X_train, y_train)

3. Analyze Influential Instances

sensitive_features = ["gender", "race"]  # Adjust based on dataset
analyzer = InfluentialInstanceAnalyzer(model, sensitive_features, deletion_percentage=10)

# Run analysis
analyzer.fit(X_train, y_train, X_test, y_test)
influential_instances, scores, acc_changes, fairness_changes = analyzer.run_analysis(showGraph=True)

4. Remove Influential Instances & Retrain

X_filtered = X.drop(index=influential_instances, errors='ignore')
y_filtered = y.drop(index=influential_instances, errors='ignore')

X_train_f, X_test_f, y_train_f, y_test_f = train_test_split(X_filtered, y_filtered, test_size=0.2, random_state=42)
model.fit(X_train_f, y_train_f)

📊 Visualization

The library supports visualizing influential instances using Seaborn:

analyzer.run_analysis(showGraph=True)

🛠 Dependencies

• NumPy
• Pandas
• Seaborn
• Matplotlib
• Scikit-learn
• Fairlearn

📝 License

This project is licensed under the MIT License.

🤝 Contributing

Contributions are welcome! Feel free to submit issues or pull requests on GitHub.

📧 Contact

Author: Blazhe Manev
GitHub: BlazheManev