pygsom - python GSOM algorithm

pygsom is an open source python-based implementation of GSOM algorithm. GSOM is unsupervised dimensionality reduction and clustering algorithm

Table of Contents

• Installation
• Minimal example
• Getting started
• [Citing gsom](#citing pygsom)

Installation

To pip install pygsom from github:

pip install pygsom

pygsom supports Python 3.6+.

Minimal example

import numpy as np
import pandas as pd
import gsom

data_filename = "data/zoo.txt".replace('\\', '/')


if __name__ == '__main__':
    np.random.seed(1)
    df = pd.read_csv(data_filename)
    print(df.shape)
    data_training = df.iloc[:, 1:17]
    gsom_map = gsom.GSOM(.83, 16, max_radius=4)
    gsom_map.fit(data_training.to_numpy(), 100, 50)
    map_points = gsom_map.predict(df,"Name","label")
    gsom.plot(map_points, "Name", gsom_map=gsom_map)
    map_points.to_csv("gsom.csv", index=False)

Getting started

Train the GSOM algorithm : need to give input data in numpy array with training iterations and smoothing iterations

gsom_map.fit(data_training.to_numpy(), <training iterations>, <smooth iterations>)

Predict cluster nodes : need to give input data in pandas dataframe with names and labels

map_points = gsom_map.predict(df,<name column name>,<label column name>)

Plot the 2D map: need to give the output of predict function with label column (name column or label column)

gsom.plot(map_points, <name column name/label column name>, gsom_map=<gsom_map>)

Citing pygsom

If you use pygsom, please cite the following paper:

@article{alahakoon2000dynamic,
  title={Dynamic self-organizing maps with controlled growth for knowledge discovery},
  author={Alahakoon, Damminda and Halgamuge, Saman K and Srinivasan, Bala},
  journal={IEEE Transactions on neural networks},
  volume={11},
  number={3},
  pages={601--614},
  year={2000},
  publisher={IEEE}
}