This is a Python Ctypes wrapper for the original SPC algorithm (available in Eytan Domany repository). The SPC code was edited to work as a shared library, reduce disk usage and improve speed and interface with Python. Two methods are included to select clusters, see Cluster Selection below. The method WC1 select clusters at one particular temperature and WC3 (default for .fit() method) selects multiple temperatures prioritizing high temperatures (this method is better to detect more clusters but a bigger number of elements will remain unlabeled).
pip install spclusteringThe original code requires gcc and makefile. For Windows I recommend TDM-GCC.
from spclustering import SPC, plot_temperature_plot
import matplotlib.pyplot as plt
import numpy as np
rng = np.random.RandomState(0)
cl1 = rng.multivariate_normal([8, 8], [[4,0],[0,3]], size=800)
cl2 = rng.multivariate_normal([0,0], [[3,0],[0,2]], size=2000)
cl3 = rng.multivariate_normal([5,5], [[0.5,0.2],[0.2,0.6]], size=300)
cl4 = rng.multivariate_normal([-3,-2.5], [[0.5,0],[0,0.6]], size=500)
gt = [cl1,cl2,cl3,cl4]
#plot clusters
plt.figure()
for cl in gt:
plt.plot(*cl.T,marker='.',linestyle='',markersize=3)
plt.title('Ground Truth')
plt.grid()
plt.show()
#run the algorithm. The fit method applied the cluster selection described in Waveclus 3.
#The method fit_WC1 is the alternative using the original Waveclus 1 temperature selection.
data = np.concatenate(gt)
clustering = SPC(mintemp=0,maxtemp=0.2)
labels, metadata = clustering.fit(data,min_clus=150,return_metadata=True)
#It is posible to show a temperature map using the optional output metadata
plot_temperature_plot(metadata)
plt.show()
#To show the assigned labels:
plt.figure()
for c in np.unique(labels):
if c==0:
plt.plot(*data[labels==c,:].T,marker='.',color='k',linestyle='',markersize=3)
else:
plt.plot(*data[labels==c,:].T,marker='.',linestyle='',markersize=3)
plt.grid()
plt.title('Results')
plt.show()
-
It runs with a data matrix [npoints, ndims] as input. Running the clustering from a distance matrix is not implemented.
-
Only the main parameters and flags are implemented.
-
Tmax (stop temperature) is not included in the temperature array (similar to Python
range). -
Included WriteEdges boolean parameter (defaul False). If True, the *.edges/*.mst/*.K file is created as in the original SPC. Added aswell PrintParam parameter in the legacy SW.c code.
-
Reduces/removes console ouputs.
Blatt, M., Wiseman, S., & Domany, E. (1996). Superparamagnetic clustering of data. Physical review letters, 76(18), 3251.
Quian Quiroga R, Nadasdy Z, Ben-Shaul Y. Unsupervised spike detection and sorting with wavelets and superparamagnetic clustering. Neural computation 16: 1661–1687, 2004.
Chaure FJ, Rey HG, Quian Quiroga R. A novel and fully automatic spike sorting implementation with variable number of features. J Neurophysiol , 2018. doi:10.1152/jn.00339.2018.
@article{spc,
title={Superparamagnetic clustering of data},
author={Blatt, Marcelo and Wiseman, Shai and Domany, Eytan},
journal={Physical review letters},
volume={76},
number={18},
pages={3251},
year={1996},
publisher={APS}
}
@article{WC1,
title = {Unsupervised spike detection and sorting with wavelets and superparamagnetic clustering},
volume = {16},
number = {8},
journal = {Neural computation},
author = {Quian Quiroga, R and Nadasdy, Zoltan and Ben-Shaul, Yoram},
year = {2004},
pages = {1661--1687},
file = {48b6995f327440100e0d7382ff2652c17c6f.pdf:I\:\\My Drive\\zotero\\storage\\GXTC9KF8\\48b6995f327440100e0d7382ff2652c17c6f.pdf:application/pdf},
}
@article{WC3,
title = {A novel and fully automatic spike sorting implementation with variable number of features},
issn = {1522-1598},
doi = {10.1152/jn.00339.2018},
journal = {Journal of Neurophysiology},
author = {Chaure, Fernando Julian and Rey, Hernan Gonzalo and Quian Quiroga, Rodrigo},
month = jul,
year = {2018},
pmid = {29995603},
keywords = {neurophysiology, single-neuron recordings, spike sorting, tetrode}
}