ClustDenFunc

This repository contains improved implementations of Fuzzy C-means (FCM), Improved Fuzzy C-means (IFCM), and K-means clustering models. These models are built using Python and leverage the pydantic library for data validation and configuration management.

📦 Installation

To install the necessary packages, use the requirements.txt file:

pip install -r requirements.txt

🚀 Usage

import pandas as pd
from fcm import FCM

# Load your data into a DataFrame
mus = [
    np.linspace(-9, -5, num=5),  
    np.linspace(-1, 0, num=5*90), 
    np.linspace(5, 7, num = 5)
]
sigs = [1, 1, 1]
grid = np.linspace(-15, 12, 10000)
Data, trueLabels = simDatabnorm(mus, sigs, grid)


# Initialize the FCM model
fcm = FCM(numClust=3, maxIter=100, mFuzzy=2.0, epsilon=1e-9, h=0.1, distance='Bhattacharyya', thetaIni='kmeans++', verbose=True)

# Train the model
fcm.fit(data)

# Predict cluster membership for new data
predictions = fcm.predict(new_data)

⚙️ Model Parameters

• numClust: Number of clusters (default: 5)
• maxIter: Maximum number of iterations (default: 150)
• mFuzzy: Fuzziness parameter for FCM and IFCM (default: 2.0)
• epsilon: Convergence threshold (default: 1e-5)
• h: Parameter for distance calculation (default: 1e-5)
• random_state: Seed for random number generation (default: None)
• verbose: Verbose output during training (default: False)
• distance: Distance metric (overlapCoef, Bhattacharyya, Hellinger, L1-distance, or a custom function)
• thetaIni: Initialization method (random, partition, overlap, or kmeans++)

🛠️ Method

• fit(Data: pd.DataFrame): Train the model with the provided data.
• predict(Data: pd.DataFrame) -> NDArray: Predict cluster membership for new data.
• defuzzication(Data: pd.DataFrame) -> NDArray: Predict the closest cluster each sample belongs to.
• partitionCoeff: Property to get the partition coefficient.
• partitionEntropyCoef: Property to get the partition entropy coefficient.

Distance Metric

• overlapCoef: Overlap coefficient distance.
• Bhattacharyya: Bhattacharyya distance.
• Hellinger: Hellinger distance.
• L1-distance: L1 distance (Manhattan distance).

📚 References

• Fuzzy C-means (FCM): J.C. Bezdek, Pattern Recognition with Fuzzy Objective Function Algorithms, Springer, 1981.
• K-means: J. MacQueen, "Some Methods for classification and Analysis of Multivariate Observations," in Proceedings of 5th Berkeley Symposium on Mathematical Statistics and Probability, 1967.