A function-based implementation of k-means clustering that maintains data association.
Project description
K-Means Clustering
A repository documenting the implementation of k-Means clustering in Python. Usage examples can be found in the tests
directory.
The thing that makes this k-means clustering module different from others is that it allows the user to specify the number of dimensions to use for the clustering operation.
For example, given some data where each element is of form
# Each element would actually be a Numpy array, but the following uses lists for readability.
[
[1, 2, 3, 4, 5],
[4, 6, 7, 8, 2],
...
]
specifying ndim=3
will result in only the first three elements of each data point being used for each operation.
This is useful for maintaining data association where it otherwise would be shuffled. An example of this is found in my implementation of image segmentation (segmentation.py
) in this same project.
Other examples of use could be for maintaining data association in object detection elements. Given some
[xmin, ymin, xmax, ymax, conf, label] # [bounding box, conf, label]
we may want to cluster the data solely on bounding box information while also maintaining the confidence intervals for each detection for further processing.
How it Works
Specifying the k
value results in a dict[int: NDArray]
where each NDArray
contains the elements within the cluster. The keys of this dict range from 0
to k-1
, allowing the key to also be used to index the corresponding cluster centroid from the centroid array.
Here is an example of the use of the cluster
function:
import numpy as np
from kmeans import cluster
data = np.random.random((10000, 7))
# Only cluster along first three elements of a given data point.
# Choose initial_means randomly (default)
clusters, centroids = cluster(data, k=4, ndim=3, tolerance=0.001)
for key in clusters:
cluster = clusters[key]
centroid = centroids[key]
print(cluster, centroid, sep="\n")
Features
- k-means clustering (no side-effects)
- k-means clustering w/ animation
- (2-D & 3-D)
- image segmentation via
kmeans.segmentation.segment_img
function
k-means Animation
Using the view_clustering
function
2-D Case (Smallest Tolerance Possible)
3-D Case (Tolerance = 0.001)
Image Segmentation
Perform image segmentation based on color groups specified by the user.
Two options:
Averaged Colors
k=4
k=10
Random Colors
k=4
Developed With
- Python (3.12.1)
- Numpy (1.26.2)
- Matplotlib (3.8.4)
However, no features specific to Python 3.12 were used.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for kmeans_tjdwill-1.0.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 1dbf7bf6d42faf357196b175cdfea797cd348c32b582b28c53868ff6b3cc83cd |
|
MD5 | c0c7f47888a6f4e0c183f5465072964f |
|
BLAKE2b-256 | 461bcec5004995504063bb9bd6e8a754c5a24ae96b9cf33cf08b2ac5f2b80221 |