sgUPCMed 0.0.1

A library for String Grammar Unsupervised Possibilistic C-Medians

What is String Grammar Fuzzy Clustering?

String Grammar Fuzzy Clustering is a clustering framework designed for syntactic or structural pattern recognition, where each data instance is represented not as a numeric vector but as a string that encodes structural information.

Unlike conventional numerical clustering method (e.g., Fuzzy C-Means), which assume that data have a fixed-length feature vector whereas structural clustering method operates directly on string data whose lengths and internal structures may vary.

In this approach, each pattern is described by a sequence of primitives (symbols) defined by grammatical rules. This is similar to how a sentence is formed from characters following syntax rules.

To measure similarity between strings, the method employs the Levenshtein distance[1], which counts the minimum number of edit operations (insertions, deletions, substitutions) required to transform on string into another.

The "fuzzy" aspect of this framework allows each string to belong to multiple clusters, with a membership degree that reflects how strongly it is associated with each cluster. This provides a more flexible and realistic clustering behavior compared to traditional "hard" clustering, which forces each sample to belong to only one group.

About This Library

This Python library introduces an algorithm belonging to the String Grammar Fuzzy Clustering framework, namely the String Grammar Unsupervised Possibilistic C-Medians (sgUPCMed).

String Grammar Unsupervised Possibilistic C-Medians (sgUPCMed)[2]

The sgUPCMed algorithm is a possibilistic clustering method for string data. The algorithm is developed based on the basic idea from the UPCM algorithm [3]. This algorithm uses the Levenshtein distance to compute distance between two strings, and the fuzzy median is utilized to calculate a cluster prototype, similar to sgFCMed [4]. Membership values are computed independently for each cluster based on the distance to the corresponding prototype, without enforcing constraints.

Key Features:

• Computes membership values as an exponential function of the distance between a string and a string prototype.
• Membership values are not constrained to sum to one across clusters for each string.
• Uses the Levenshtein distance to measure dissimilarity between strings.
• Represents each cluster prototype using a fuzzy median string.
• Suitable for unsupervised clustering of unlabeled string data with varying lengths.

**Please be noted that this sgUPCMed can be used for academic and research purposes only. Please also cite this paper [2].**

Reference

[1] S. K. Fu, Syntactic Pattern Recognition and Applications, 1982, Prentice-Hall, Zbl0521.68091.

[2] Atcharin Klomsae, Sansanee Auephanwiriyakul, and Nipon Theera-Umpon. “A Novel String Grammar Unsupervised Possibilistic C-Medians Algorithm for Sign Language Translation Systems”, Symmetry (special issue: Emerging Approaches and Advances in Big Data), Vol. 9, No. 12, December 2017.

[3] Yang, M.S. and Wu, K.L., “Unsupervised Possibilistic Clustering,” Pattern Recognition, Vol 39, pp. 5-21, 2006.

[4] Atcharin Klomsae, Sansanee Auephanwiriyakul, and Nipon Theera-Umpon, “A Novel String Grammar Fuzzy C-Medians,” Proceedings of the 2015 IEEE International Conference on Fuzzy Systems, Istanbul, Turkey, August 2015.

Installation

You can install the library using pip:

pip install sgUPCMed

USAGE

Example Code

import random
from sgUPCMed import SGUPCMed # Import the clustering class

if __name__ == "__main__":
    # Set random seed for reproducibility
    random.seed(42)

    # Define a list of strings to cluster
    data = ["book", "back", "boon", "cook", "look", "cool", "kick", "lack", "rack", "tack"]

    # Create the model with 2 clusters and fuzzifier m=2.0
    model = SGUPCMed(C=2, m=2.0)

    # Fit the model on the data
    model.fit(data)

    # Print the final prototype strings representing each cluster
    print("Prototypes:", model.prototypes())

    # Print the fuzzy membership matrix for each input string
    print("\nMembership Matrix (U):")
    for s, u in zip(data, model.membership()):
        print(f"{s:>6} → {[val for val in u]}")

    # Define new strings to classify using the trained model
    new_data = ["hack", "rook", "cook"]

    # Predict the cluster index (0 or 1) for each new string
    preds = model.predict(new_data, model.prototypes())
    print("\nPredictions:")
    for s, c in zip(new_data, preds):
        print(f"{s} → Cluster {c+1}")