Guide¶
Installation¶
This project is distributed as a Python package and is hosted on the PyPI package server. To use cvi, first install it using pip:
pip install cvi
You can also add the package directly from GitHub to get the latest changes between releases (or from a specific branch) with:
pip install git+https://github.com/AP6YC/cvi
Quickstart¶
This section provides a quick overview of how to use the project. For more detailed code usage, please see the Detailed Usage section.
Create a CVI object and compute the criterion value in batch with get_cvi:
# Import the library
import cvi
# Create a Calinski-Harabasz (CH) CVI object
my_cvi = cvi.CH()
# Load some data from some clustering algorithm
samples, labels = load_some_clustering_data()
# Compute the final criterion value in batch
criterion_value = my_cvi.get_cvi(samples, labels)
or do it incrementally, also with get_cvi:
# Datasets are numpy arrays
import numpy as np
# Create a container for criterion values
n_samples = len(labels)
criterion_values = np.zeros(n_samples)
# Iterate over the data
for ix in range(n_samples):
criterion_values = my_cvi.get_cvi(samples[ix, :], labels[ix])
Detailed Usage¶
The cvi package contains a set of implemented CVIs with batch and incremental update methods. Each CVI is a standalone stateful object inheriting from a base class CVI, and all CVI functions are object methods, such as those that update parameters and return the criterion value.
Instantiate a CVI of you choice with the default constructor:
# Import the package
import cvi
# Import numpy for some data handling
import numpy as np
# Instantiate a Calinski-Harabasz (CH) CVI object
my_cvi = cvi.CH()
CVIs are instantiated with their acronyms, with a list of all implemented CVIS being found in the [Implemented CVIs](#implemented-cvis) section.
A batch of data is assumed to be a numpy array of samples and a numpy vector of integer labels.
# Load some data
samples, labels = my_clustering_alg(some_data)
Note
The cvi package assumes the Numpy row-major convention where rows are individual samples and columns are features. A batch dataset is then [n_samples, n_features] large, and their corresponding labels are [n_samples] large.
You may compute the final criterion value with a batch update all at once with CVI.get_cvi
# Get the final criterion value in batch mode
criterion_value = my_cvi.get_cvi(samples, labels)
or you may get them incrementally with the same method, where you pass instead just a single numpy vector of features and a single integer label. The incremental methods are used automatically based upon the dimensions of the data that is passed.
# Create a container for the criterion value after each sample
n_samples = len(labels)
criterion_values = np.zeros(n_samples)
# Iterate across the data and store the criterion value over time
for ix in range(n_samples):
sample = samples[ix, :]
label = labels[ix]
criterion_values[ix] = my_cvi.get_cvi(sample, label)
Note
After batch initialization, additional samples may be added incrementally by passing a single sample and label to get_cvi.
Remove and Merge¶
An initialized CVI can remove a previously added sample or merge two existing clusters without retaining and replaying the full dataset:
# Remove a sample from its current cluster.
criterion_value = my_cvi.remove(sample, label)
# Merge every member of source_label into target_label.
criterion_value = my_cvi.merge(target_label, source_label)
Both methods update the object in place and return its new criterion value. Removing the final sample of a cluster deletes that cluster, while merge retains target_label and deletes source_label. The caller is responsible for ensuring that a removed sample belongs to the supplied label.
Add, remove, and merge are supported after either incremental or batch initialization.
Implemented CVIs¶
The following CVIs have been implemented as of the latest version of cvi:
CH: Calinski-Harabasz
cSIL: Centroid-based Silhouette
DB: Davies-Bouldin
GD43: Generalized Dunn’s Index 43.
GD53: Generalized Dunn’s Index 53.
PS: Partition Separation.
rCIP: (Renyi’s) representative Cross Information Potential.
WB: WB-index.
XB: Xie-Beni.
Acknowledgements¶
Derivation¶
The incremental and batch CVI implementations in this package are largely derived from the following Julia language implementations by the same authors of this package: