ExplodingHam

Oddball scikit-learn style ML+ approaches.

This library provides scikit-learn style solutions for ML problems. The goal of this library is to extend beyond what scikit-learn to include obscure ML approaches, as well as non-ML approaches that you can easily baseline your ML approachs against.

Models

A couple examples of currently available models include:

• Normalized Compression Distance KNN, this is a nifty approach to building a classifier on text data.
• RegEx Classifier, takes a provided RegEx and uses it to create a quick and dirty classifier. This is not machine learning, it doesn't learn, but having this as a scikit-learn classifier allows you to plug and play with your existing code to provide a baseline, or experiment with different versions of your RegEx.

Future methods to include:

• Bumping Classifier, an ensemble method that trains many models on subsets of the data, and keeps the model that performs the best.
• Rotation Forests, a variation on a random forest that performs PCA before each tree is fit.
• InfoGain Trees, classifiers that use ID3 and C4.5 on non-binary trees.
• Zero-Shot LLM Classifier, a classifier that uses an LLM with a prompt to make predictions.
• Few-Shot LLM Classifier, a classifier that uses an LLM with multiple examples to provide a classification prediction.
• Normalized Compression Distance Clustering, hierarchical clustering, KMeans, DBScan using compression algorithms to represent the data.

Integrations:

• ExplodingHam uses the narwhals library for dataset processing, this means that it is plug and play with most DataFrames you might supply.
• ExplodingHam uses base classes from scikit-learn, so these models should be relatively indistinguishable from