![Input data format for BANG
BANG algorithm uses array_like data format, for example,
list where each element is a point that is represented by
coordinates. Here is an example of 1-D data:
dataset = [ [0.25], [0.43], [1.34], [-0.56] ]
Example of 2-D data:
dataset = [ [0.1, 0.2], [0.1, 0.6], [0.2, -1.4] ]
Example of 3-D data:
dataset = [ [0.6, 0.1, 0.4], [0.2, 0.1, 0.9] ]](https://image.slidesharecdn.com/pyclusteringtutorial-bang-180602173212/85/Pyclustering-tutorial-BANG-5-320.jpg)
![Clustering by BANG algorithm
BANG uses one general parameter for clustering: level of
division that technically means amount of BANG tree hight
where root is a cell that covers whole data.
from pyclustering.cluster.bang import bang
data = [ [0.1], [0.2], [0.5], [0.3], [1.5], [1.8], [1.3], [1.6], [1.5] ]
levels = 3
instance = bang(data, levels)
instance.process() # perform processing
clusters = instance.get_clusters(); # allocated clusters
noise = instance.get_noise(); # cluster noise
directory = instance.get_directory(); # BANG directory
dendrogram = instance.get_dendrogram() # BANG dendrogram](https://image.slidesharecdn.com/pyclusteringtutorial-bang-180602173212/85/Pyclustering-tutorial-BANG-6-320.jpg)
Pyclustering tutorial - BANG
- 1. PyClustering: BANG Tutorial Pyclustering Library Tutorial Theme: BANG Library Version: 0.8.2 Author: Andrei Novikov E-Mail: [email protected]
- 2. Agenda ● PyClustering BANG Features; ● Input format for BANG; ● Data clustering using BANG; ● Data clustering result visualization; ● Animation of BANG clustering process; ● Image segmentation by BANG;
- 3. PyClustering BANG Features ● BANG algorithm provides clustering results that can be used for further analysis (not only clusters and noise itself): dendrogram, BANG directory - information about density distribution, structure, approximate amount of real clusters and influence of input argument on clustering results. ● BANG visualizer for clustering results, BANG directory and dendrogram. ● BANG animator for creating animation of BANG clustering process.
- 4. Import BANG BANG algorithm and its features can be imported from „pyclustering.cluster.bang“: from pyclustering.cluster.bang import bang from pyclustering.cluster.bang import bang_directory from pyclustering.cluster.bang import bang_visualizer from pyclustering.cluster.bang import bang_animator
- 5. Input data format for BANG BANG algorithm uses array_like data format, for example, list where each element is a point that is represented by coordinates. Here is an example of 1-D data: dataset = [ [0.25], [0.43], [1.34], [-0.56] ] Example of 2-D data: dataset = [ [0.1, 0.2], [0.1, 0.6], [0.2, -1.4] ] Example of 3-D data: dataset = [ [0.6, 0.1, 0.4], [0.2, 0.1, 0.9] ]
- 6. Clustering by BANG algorithm BANG uses one general parameter for clustering: level of division that technically means amount of BANG tree hight where root is a cell that covers whole data. from pyclustering.cluster.bang import bang data = [ [0.1], [0.2], [0.5], [0.3], [1.5], [1.8], [1.3], [1.6], [1.5] ] levels = 3 instance = bang(data, levels) instance.process() # perform processing clusters = instance.get_clusters(); # allocated clusters noise = instance.get_noise(); # cluster noise directory = instance.get_directory(); # BANG directory dendrogram = instance.get_dendrogram() # BANG dendrogram
- 7. Output result of BANG BANG algorithm returns allocated clusters, outliers (noise), BANG directory and dendrogram: ● Clusters are represented by list of clusters and each cluster contains object indexes from dataset. ● Outliers are represented by list of object indexes from dataset. ● BANG directory is a object of class bang_directory. ● Dendrogram is represented by list of dendrogram clusters where each cluster contains expand order of blocks that was formed during clustering process.
- 8. BANG result visualization Output result such as clusters and outliers can be shown by common visualizer „cluster_visualizer“ from „pyclustering.cluster“ or by BANG visualizer: instance = bang(sample, levels) instance.process() clusters = instance.get_clusters() # allocated clusters noise = instance.get_noise() # outliers directory = instance.get_directory() # get BANG directory # visualize clustering results bang_visualizer.show_clusters(sample, clusters, noise) bang_visualizer.show_dendrogram(dendrogram) bang_visualizer.show_blocks(directory)
- 9. BANG visualization – Lsun Visualization of BANG clustering results. Data: ‚Lsun‘. BANG blocks and their densities BANG dendrogram of three formed clusters Allocated clusters by BANG algorithm
- 10. BANG visualization – Chainlink BANG visualizer is able to display clustering result of N- dimension data by forming pair xk, xl, where k, l – dimension indexes. BANG blocks representation in case 3-dimensional data
- 11. BANG visualization - Target Visualization of BANG clustering results. Data: ‚Target‘. BANG blocks and their densities BANG dendrogram of three formed clusters Allocated clusters by BANG algorithm
- 12. Animate BANG clustering process BANG clustering process can be animated (in case of 2- dimensional data) using services of class ‚bang_animator‘. instance = bang(data, levels, density_threshold) instance.process() # obtain BANG clustering result for creating animation clusters = instance.get_clusters() noise = instance.get_noise() directory = instance.get_directory() # show animation bang_animator(directory, clusters, noise).animate() # or save animation to file bang_animator(directory, clusters, noise).animate(movie_filename=bang.mp4)
- 13. BANG for segmentation BANG can consider image as a data where each point represent pixel with three coordinate (RGB), or with four coordinate (RGBA). Here is an example of image segmentation using BANG algorithm: from pyclustering.utils import draw_image_mask_segments, read_image from pyclustering.cluster.bang import bang; data = read_image(file_path_to_image); bang_instance = bang(data, levels, density_threshold); bang_instance.process(); clusters = bang_instance.get_clusters(); draw_image_mask_segments(source, clusters);
- 14. Image segmentation results Example of image segmentation of photo from St.- Petersburg by BANG algorithm using pyclustering. Shortcoming of BANG segmentation is not trivial way to set density Threshold, especially to handle Images with color gradient
- 15. References and Links ● Official pyclustering github repository: https://github.com/annoviko/pyclustering ● Official pypi pyclustering page: https://pypi.org/project/pyclustering/ ● Official pyclustering web-site: https://pyclustering.github.io/
- 16. Thank you for your attention Thank you!