Marvin

A small machine learning library in Haskell. My Bachelor's thesis.

Parts of the design was inspired by the popular scikit-learn Python library.

Features

• Type classes for machine learning tasks: fit, predict, evaluate.
• Combining transformations with Pipeline type, an Arrow-instance. Instead of using transformers as in scikit-learn, we make preprocessing composable with a pinch of category theory.
• Error handling with Either.
• Table: column-oriented representation of datasets (also for validating data type).
• Small (and ugly) GUI for showcasing.

How to build

Use stack. Recommended to install gtk-buildtools before building.

Documentation

See haddock. Generate them with e.g. stack haddock.

Datasets for running examples

You can run the examples with publicly available datasets.

• HousingPrices: https://archive.ics.uci.edu/ml/datasets/Housing
• NetworkIntrusionDetection: http://kdd.ics.uci.edu/databases/kddcup99/kddcup99.html
• SpamClassification: http://www2.aueb.gr/users/ion/data/enron-spam/