Overview

Dungeons and Deep Learning uses various applications from the realm of fantasy tabletop role-playing games to educate about topics in machine learning in an approachable setting. Starting with the classification of names into the different ancestries, simple concepts involving manually crafted features can be taught and implemented from scratch. Via more sofisticated deep-learning approaches, eventually, generative models are used to create new random content, such as names given their ancestry.

The goal is educational as much as technical: build each stage visibly, explain why it exists, and connect abstract machine-learning concepts to a result that is easy to inspect.

Current Scope

The project currently focuses on the data foundation, manual feature engineering, and implementation of classical machine learning algorithms from scratch. It can collect categorized fantasy names, normalize and load text data, and calculate string-level features. Tests cover the reusable feature and input/output utilities. The first methods to be implemented are decision trees. The concepts are taught through Jupyter notebooks with markdown cells for explanations and code cells to fill in solutions.

Learning Roadmap

  1. Data exploration: build a categorized corpus and discuss provenance, quality, and bias.
  2. Feature engineering: turn variable-length names into numerical features or character sequences.
  3. Baselines: establish simple probabilistic generators before adding neural networks.
  4. Modeling: compare sequence architectures and training behavior.
  5. Evaluation: combine quantitative checks with human judgment of novelty and style.

The project remains active and exploratory; a trained name-generation model is not yet part of the repository.