Sign in to view source links and access this dataset
Description
613,399 No-Limit Hold'em hands in Open Hand History format (spec 1.4.7) were generated via self-play using rs-poker's arena. The dataset, created by otter-crew, serves as the training set for the range-reader model, which predicts a villain's hole cards from the action. It was last updated on June 15, 2026.
Use Cases
Train models to predict opponent hole cards based on the action sequences described in the hand histories.
Benchmark poker AI agents using a corpus of simulated cash game hands.
Analyze poker strategy and player tendencies from the structured hand records.
Develop and test hand-range estimation algorithms for No-Limit Hold'em.
Strengths
Contains 613,399 individual poker hands, providing a substantial corpus for model training.
Hands are structured in the standardized Open Hand History format (spec 1.4.7), suggesting consistency.
Hole cards are shown for every seat, which is a key feature for range prediction tasks.
Limitations
Description metadata is limited; actual data quality requires manual inspection after download.
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Provenance
Source
Self-play from rs-poker's arena.
Collection Method
Generated via self-play simulation for cash games with 5/10 blinds, randomized stacks, and 2- to 6-handed tables.
Freshness
Last updated 2026-06-15 16:05:29; freshness should be verified.