Datasets:
The dataset viewer is not available for this dataset.
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
YAML Metadata Warning:The task_categories "tabular" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
Dataset Card for ARC-AGI 3 Public Demo Human Testing
Dataset Summary
This dataset contains human gameplay logs and trajectories from the ARC-AGI 3 public demo. It is a fully open-source dataset created by the ARC Prize.
The primary purpose of publishing this dataset on Hugging Face is to make it easily accessible and convenient for participants in the Kaggle ARC Prize 2026 Competition.
The implementation and source code used to process and upload this dataset to Hugging Face can be found in this Kaggle Notebook: ARC-AGI 3 Public Demo Human Testing (Hugging Face).
Dataset Structure
The dataset consists of a single train split with 340 rows. Each row represents a single game session/attempt by a human user.
Data Fields
- env: The environment identifier (e.g.,
sk48). - guid: A unique identifier for the user's recording session.
- trajectory: A JSON string capturing the chronological sequence of events, timestamps, and game-specific data.
- timestamp:str The timestamp when environmental observation data was returned in response to the action. (Probably)
- data:arcengine.FrameDataRaw Observational data returned from the environment by the action The data structure of FrameDataRaw has been confirmed as follows: https://www.kaggle.com/competitions/arc-prize-2026-arc-agi-3/discussion/705068
- result_timestamp: The timestamp when the session ended.
- won:
1if the user successfully completed the game/levels,0otherwise. - played: Number of games played in this session.
- total_actions: Total number of actions/moves taken by the player.
- levels_completed: A list indicating which levels within the session were successfully solved (e.g.,
[8]). - game_id: The specific game instance ID.
- states: The final state of the game (e.g.,
WIN,GAME_OVER,NOT_FINISHED). - actions_by_level: A nested list breaking down the action counts at each specific level.
- resets: The number of times the player reset the puzzle/level.
Intent and Use Case
By utilizing this dataset to perform imitation learning on human problem-solving approaches, we believe it can significantly help AI agents learn effective action planning and reasoning strategies on the ARC benchmark. It serves as a valuable resource for Kaggle participants aiming to bridge the gap between human and machine intelligence.
- Downloads last month
- 223