A large-scale StarCraft: Brood War replay dataset for AI research, containing 65,646 games with frame and action data.
StarData is a large-scale dataset of StarCraft: Brood War replays created for AI and machine learning research. It provides frame-by-frame game data and player actions to train and evaluate AI agents in complex real-time strategy environments. The dataset addresses the need for standardized, high-quality benchmarks in game AI research.
AI researchers and machine learning practitioners working on real-time strategy games, reinforcement learning, or game AI benchmarking. It's particularly valuable for those developing agents for StarCraft or similar complex environments.
Researchers choose StarData because it offers the largest publicly available StarCraft replay dataset with carefully curated, high-quality data and multi-language accessibility through TorchCraft. Its standardized splits and reproducibility tools make it ideal for benchmarking and comparative research.
Starcraft AI Research Dataset
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
With 65,646 games and 1.535 billion frames, it offers one of the largest collections of StarCraft replays, enabling extensive training data for AI models.
Replays are cleaned and selected to ensure high quality, reducing corrupt data and improving research reliability as noted in the README.
Accessible via TorchCraft in C++, Python, and Lua, providing flexibility for researchers using different programming environments.
Includes standardized train/validation/test splits and scripts for stats extraction and battle clustering, facilitating reproducible experiments.
Tied to TorchCraft version 1.3.0, which is old and may not be actively maintained, posing compatibility issues as stated in the README.
The compressed dataset is 365 GB, requiring significant storage space that might be prohibitive for individual researchers or small teams.
Installation involves managing git submodules, specific library versions like libzstd-1.1.4+, and multiple components, adding to setup time and complexity.
Exclusively for StarCraft: Brood War, limiting its applicability to other games or broader AI domains without adaptation.