lshkrepresentatives 1.2.0

LSH-k-Representatives: A categorial clustering algorithm using LSH

Python implementations of the LSH-k-Representatives algorithms for clustering categorical data:

Different from k-Modes algorithm, LSH-k-Representatives define the "representatives" that keep the frequencies of all categorical values of the clusters.

Notebook Samples:

Applying LSHkRepresentatives:

https://github.com/nmtoan91/lshkrepresentatives/blob/main/notebook_sample.ipynb

Applying LSHkRepresentatives_Full (for huge dataset):

https://github.com/nmtoan91/lshkrepresentatives/blob/main/notebook_sample_LSHkRepresentatives_Full.ipynb

Normalizing unstructed normal dataset:

https://github.com/nmtoan91/lshkrepresentatives/blob/main/notebook_dataset_normalization.ipynb

Note: The dataset is auto normalized if it detect string, or disjointed data, or nan

Installation:

Using pip:

pip install lshkrepresentatives numpy scikit-learn pandas kmodes networkx termcolor

Import the packages:

import numpy as np
from LSHkRepresentatives.LSHkRepresentatives import LSHkRepresentatives

Generate a simple categorical dataset:

X = np.array([['red',0,np.nan],['green',1,1],['blue',0,0],[1,5111,1],[2,2,2],[2,6513,'rectangle'],[2,3,6565]])

LSHk-Representatives (Init):

#Init instance of LSHkRepresentatives 
kreps = LSHkRepresentatives(n_clusters=2,n_init=5) 
#Do clustering for dataset X
labels = kreps.fit(X)
#Print the label for dataset X
print('Labels:',labels)
#Predict label for the random instance x
x = np.array(['red',5111,0])
label = kreps.predict(x)
print(f'Cluster of object {x} is: {label}')

Outcome:

SKIP LOADING distMatrix because: False bd=None
Generating disMatrix for DILCA
Saving DILCA to: saved_dist_matrices/json/DILCA_None.json
Generating LSH hash table:   hbits: 2(4)  k 1  d 3  n= 7
LSH time: 0.006518099999993865 Score:  6.333333333333334  Time: 0.0003226400000130525
Labels: [1 1 1 1 0 0 0]
Cluster of object [1 2 0] is: 1

Built-in evaluattion metrics:

y = np.array([0,0,0,0,1,1,1])
kreps.CalcScore(y)

Outcome:

Purity: 1.00 NMI: 1.00 ARI: 1.00 Sil:  0.59 Acc: 1.00 Recall: 1.00 Precision: 1.00

LSHk-Representatives (Full):

This version of LSHk-Representatives target for huge dataset, the accuracy will be reduced but the speed is increase from 2 to 32 times depend on the data

X = np.array([[0,0,0],[0,1,1],[0,0,0],[1,0,1],[2,2,2],[2,3,2],[2,3,2]])
kreps = LSHkRepresentatives_Full(n_clusters=2,n_init=5) 
labels = kreps.fit(X)
print('Labels:',labels)
x = np.array([1,2,0])
label = kreps.predict(x)
print(f'Cluster of object {x} is: {label}')

Built-in evaluattion metrics:

y = np.array([0,0,0,0,1,1,1])
kreps.CalcScore(y)

Out come:

SKIP LOADING distMatrix because: True bd=None
Generating disMatrix for DILCA
Saving DILCA to: saved_dist_matrices/json/DILCA_None.json
Generating LSH hash table:   hbits: 2(4)  k 2  d 3  n= 7
 n_group=2 Average neighbors:1.0
LSH time: 0.00661619999999985 Score:  6.333333333333334  Time: 0.000932080000000024
Purity: 1.00 NMI: 1.00 ARI: 1.00 Sil:  0.59 Acc: 1.00 Recall: 1.00 Precision: 1.00

Parameters:

X: Categorical dataset
y: Labels of object (for evaluation only)
n_init: Number of initializations
n_clusters: Number of target clusters
max_iter: Maximum iterations
verbose:
random_state:

If the variable MeasureManager.IS_LOAD_AUTO is set to "True": The DILCA will get the pre-caculated matrix

Outputs:

cluster_representatives: List of final representatives
labels_: Prediction labels
cost_: Final sum of squared distance from objects to their centroids
n_iter_: Number of iterations
epoch_costs_: Average time for an initialization

References:

T. N. Mau and V.-N. Huynh, ``An LSH-based k-Representatives Clustering Method for Large Categorical Data." Neurocomputing, Volume 463, 2021, Pages 29-44, ISSN 0925-2312, https://doi.org/10.1016/j.neucom.2021.08.050.

Bibtex:

@article{mau2021lsh,
  title={An LSH-based k-representatives clustering method for large categorical data},
  author={Mau, Toan Nguyen and Huynh, Van-Nam},
  journal={Neurocomputing},
  volume={463},
  pages={29--44},
  year={2021},
  publisher={Elsevier}
}

pypi/github repository

https://pypi.org/project/lshkrepresentatives/
https://github.com/nmtoan91/lshkrepresentatives