How to use Scikit-learn's affinity propagation clustering with my own datasets?

Question

I am trying to cluster my datasets using affinity propagation. I followed this and this links to grasp the basics of affinity propagation clustering. The sample code available at sklearn is as follows:

from sklearn.cluster import AffinityPropagation
from sklearn import metrics
from sklearn.datasets.samples_generator import make_blobs
# Generate sample data
centers = [[1, 1], [-1, -1], [1, -1]]
X, labels_true = make_blobs(n_samples=300, centers=centers, cluster_std=0.5,
                            random_state=0)
# Compute Affinity Propagation
af = AffinityPropagation(preference=-50).fit(X)
cluster_centers_indices = af.cluster_centers_indices_
labels = af.labels_

While running it, it works as stated on the website. However, I could not understand it completely. I want to modify this code to use with my datasets. My datasets consist of values from different sensors on a 2-D surface. I want to cluster the values with similar sensors readings at various points on the 2-D surface. How can I do it?

Thank you.

Answer 1

Clustering algorithms usually assume that you have positions of objects, and want to find dense groups of observations.

If I understand you correctly, you have a 2d grid of sensor readings, and you want to segment them into regions. That is a slightly different problem. If you'd just put your sensor readings into a clustering, then the clusters will not be spatially coherent: clustering assumes there is no particular order to the points.

So you'll need to look into segmentation.

A naive way would be to use (sensor.x, sensor.y, sensor.value) tuples. Including the sensor positions will cause the results to be somewhat spatially coherent. But that makes it very sensitive to scaling, and there is no "correct" way of scaling this. There is a trade-off between spatial coherence and measurement coherence.