OmenMachine

A content based recommendation system for Magic: The Gathering cards using scikit-learn. The relevant input for our machine-learning algorithm are the so-called features. These represent a set of card properties that we use to compare it to other cards. Examples of interesting features are:

• The oracle text, meaning a description of what the card does.
• The converted mana cost to play the card.
• The type of the card (e.g., Creature or Instant).

The combined features are converted into a count matrix. Based on this matrix, we compute the cosine similarity, which is stored as a correlation matrix. The output is a ranking of cards that are most similar to the input card in terms of similarity score. Additional filters, for example color identity or legality in various formats can be applied.

Graphical user interface

GUI
Start screen
Selection
Results screen

Requirements

• Numpy
• Pandas
• JSON
• joblib
• scikit-learn
• re

Optional requirements for GUI

• tkinter
• urllib.parse
• PIL
• io
• webbrowser

Copyright

The used information about Magic: The Gathering cards are compiled by Scryfall (https://scryfall.com) and is copyrighted by Wizards of the Coast, LLC, a subsidiary of Hasbro, Inc. This project is permitted under the Wizards of the Coast Fan Content Policy.