Predicting Chess Puzzle Difficulty – IEEE BigData 2024 Cup (4th Place)

This repository contains my submission to the IEEE BigData 2024 Cup: Predicting Chess Puzzle Difficulty, where I placed 4th.

The solution is based on a learning-to-rank approach.

Contents

• glicko_cnn.ipynb: Notebook used to train the model
• glicko_inference.ipynb: Notebook for generating predictions on test data

Approach

The core idea is to frame puzzle difficulty prediction as a pairwise ranking problem, to predict the relative difficulty between puzzle pairs rather than absolute scores.

You can find a detailed description of the approach in the following resources:

• 📄 Paper on IEEE Xplore
• 📄 Paper (PDF)
• 🎤 Presentation slides

Related Work

• 🥇 Code of the 1st place winning solution