Kaushik Rajan
Phase 1: Initial SPIRAL project setup
e526e6a
|
raw
history blame
1.32 kB

SPIRAL Data Directory

This directory contains datasets, benchmarks, and cached data for the SPIRAL Interactive Reasoning Game Simulator.

Structure

data/
β”œβ”€β”€ cache/              # Cached model outputs and processed data
β”œβ”€β”€ datasets/           # Game datasets and training data
β”œβ”€β”€ benchmarks/         # Evaluation benchmarks for transfer learning
β”‚   β”œβ”€β”€ gsm8k.json     # GSM8K math problems
β”‚   └── logic_puzzles.json  # Logic reasoning puzzles
└── README.md          # This file

Datasets

Game Datasets

  • Kuhn Poker: Training games and strategies
  • TicTacToe: Game states and optimal moves

Benchmark Datasets

  • GSM8K: Grade School Math 8K dataset for mathematical reasoning
  • Logic Puzzles: Custom logic and reasoning problems
  • Strategic Reasoning: Game-theory based reasoning tasks

Usage

Datasets are automatically downloaded and cached when first used. To manually download:

from src.data_utils import download_datasets
download_datasets()

Data Sources

  • GSM8K: Cobbe et al. 2021
  • Logic Puzzles: Curated collection from various sources
  • Game Data: Generated through self-play training

License

Please refer to individual dataset licenses for usage rights.