This module implements a K-Nearest Neighbors classifier that uses Normalized Compression Distance (NCD) to measure similarity between instances. The class KNN_NCD provides methods for fitting the model to the training data and predicting the class of test instances.
Author: Asher Bond
Contact: source@elasticprovisioner.com / @AsherBond on GitHub
Ensure that you have gzip and numpy installed in your Python environment. You can install them using pip:
pip install numpyInstall various compression algorithms if you're using one not built into Python.
pip install zstandard python-snappy lz4The classifier uses a compression algorithm to compute the Normalized Compression Distance. By default, it uses the gzip algorithm. However, you can use a different algorithm by passing its name as a string to the KNN_NCD constructor, like so:
knn = KNN_NCD(k=1, compression='lzma')The currently supported compression algorithms are: "gzip", "bz2", "lzma", "zstd", "snappy", and "lz4".
These compression algorithms are built into Python:
- gzip
- bz2
- lzma
These compression algorithms require the following Python libraries:
- zstd: Install with
pip install zstandard - snappy: Install with
pip install python-snappy - lz4: Install with
pip install lz4
If you try to use a compression algorithm without the required library installed, or if you specify a compression algorithm that is not supported, the KNN_NCD constructor will raise a ValueError.
Let's say you have a classification task where you need to classify text strings into different categories (for example, spam and not-spam). Here's how you can use the NCD Classifier:
from knn_ncd import KNN_NCD
# Initialize training data
training_data = [("spam text", "spam"), ("not spam text", "not-spam"), ...]
# Initialize test data
test_data = ["this is a spam text", "this is not spam", ...]
# Create an instance of KNN_NCD with k as 3
knn_ncd = KNN_NCD(k=3)
# Train the model
knn_ncd.fit(training_data)
# Predict the class labels for test data
predictions = knn_ncd.predict(test_data)In this case, the knn_ncd object is trained to classify text strings as "spam" or "not-spam". The number of neighbors k is set to 3, meaning the classifier will consider the 3 nearest neighbors according to the Normalized Compression Distance to make its prediction.
You can run the tests for this project using Python's built-in unittest module. Navigate to the project directory and run the following command:
python -m unittest test_knn_ncd.py
The tests check the functionality of the NCD calculation and the prediction mechanism.
Feel free to fork the project, open a pull request, or submit suggestions and bugs in the issue tracker.
Apache 2.0 License is included (LICENSE.txt)