| { |
| "metadata": { |
| "description": "Distribution of QA pairs across domains and problem types", |
| "total_episodes": 208, |
| "total_qa_pairs": 2496, |
| "problem_type_mapping": { |
| "A": "Recall", |
| "B": "Causal Inference", |
| "C": "State Updating", |
| "D": "State Abstraction" |
| } |
| }, |
| "overall_distribution": { |
| "total_qa_pairs": 2496, |
| "problem_types": { |
| "A": { |
| "name": "Recall", |
| "count": 839, |
| "ratio": 0.336058 |
| }, |
| "B": { |
| "name": "Causal Inference", |
| "count": 596, |
| "ratio": 0.238782 |
| }, |
| "C": { |
| "name": "State Updating", |
| "count": 647, |
| "ratio": 0.259296 |
| }, |
| "D": { |
| "name": "State Abstraction", |
| "count": 414, |
| "ratio": 0.165865 |
| } |
| } |
| }, |
| "domain_distribution": { |
| "TEXT2SQL": { |
| "total_episodes": 51, |
| "episode_ratio": 0.245192, |
| "total_qa_pairs": 612, |
| "qa_ratio": 0.245192, |
| "problem_types": { |
| "A": { |
| "name": "Recall", |
| "count": 223, |
| "ratio_in_domain": 0.364379, |
| "ratio_overall": 0.089344 |
| }, |
| "B": { |
| "name": "Causal Inference", |
| "count": 153, |
| "ratio_in_domain": 0.250000, |
| "ratio_overall": 0.061298 |
| }, |
| "C": { |
| "name": "State Updating", |
| "count": 134, |
| "ratio_in_domain": 0.218954, |
| "ratio_overall": 0.053686 |
| }, |
| "D": { |
| "name": "State Abstraction", |
| "count": 102, |
| "ratio_in_domain": 0.166667, |
| "ratio_overall": 0.040865 |
| } |
| } |
| }, |
| "SOFTWARE": { |
| "total_episodes": 36, |
| "episode_ratio": 0.173077, |
| "total_qa_pairs": 432, |
| "qa_ratio": 0.173077, |
| "problem_types": { |
| "A": { |
| "name": "Recall", |
| "count": 212, |
| "ratio_in_domain": 0.490741, |
| "ratio_overall": 0.084936 |
| }, |
| "B": { |
| "name": "Causal Inference", |
| "count": 75, |
| "ratio_in_domain": 0.173611, |
| "ratio_overall": 0.030048 |
| }, |
| "C": { |
| "name": "State Updating", |
| "count": 73, |
| "ratio_in_domain": 0.168981, |
| "ratio_overall": 0.029247 |
| }, |
| "D": { |
| "name": "State Abstraction", |
| "count": 72, |
| "ratio_in_domain": 0.166667, |
| "ratio_overall": 0.028846 |
| } |
| } |
| }, |
| "WEB": { |
| "total_episodes": 31, |
| "episode_ratio": 0.149038, |
| "total_qa_pairs": 372, |
| "qa_ratio": 0.149038, |
| "problem_types": { |
| "A": { |
| "name": "Recall", |
| "count": 125, |
| "ratio_in_domain": 0.336022, |
| "ratio_overall": 0.050080 |
| }, |
| "B": { |
| "name": "Causal Inference", |
| "count": 93, |
| "ratio_in_domain": 0.250000, |
| "ratio_overall": 0.037260 |
| }, |
| "C": { |
| "name": "State Updating", |
| "count": 93, |
| "ratio_in_domain": 0.250000, |
| "ratio_overall": 0.037260 |
| }, |
| "D": { |
| "name": "State Abstraction", |
| "count": 61, |
| "ratio_in_domain": 0.163978, |
| "ratio_overall": 0.024439 |
| } |
| } |
| }, |
| "GAME": { |
| "total_episodes": 30, |
| "episode_ratio": 0.144231, |
| "total_qa_pairs": 360, |
| "qa_ratio": 0.144231, |
| "problem_types": { |
| "A": { |
| "name": "Recall", |
| "count": 120, |
| "ratio_in_domain": 0.333333, |
| "ratio_overall": 0.048077 |
| }, |
| "B": { |
| "name": "Causal Inference", |
| "count": 90, |
| "ratio_in_domain": 0.250000, |
| "ratio_overall": 0.036058 |
| }, |
| "C": { |
| "name": "State Updating", |
| "count": 90, |
| "ratio_in_domain": 0.250000, |
| "ratio_overall": 0.036058 |
| }, |
| "D": { |
| "name": "State Abstraction", |
| "count": 60, |
| "ratio_in_domain": 0.166667, |
| "ratio_overall": 0.024038 |
| } |
| } |
| }, |
| "EMBODIED_AI": { |
| "total_episodes": 30, |
| "episode_ratio": 0.144231, |
| "total_qa_pairs": 360, |
| "qa_ratio": 0.144231, |
| "problem_types": { |
| "A": { |
| "name": "Recall", |
| "count": 61, |
| "ratio_in_domain": 0.169444, |
| "ratio_overall": 0.024439 |
| }, |
| "B": { |
| "name": "Causal Inference", |
| "count": 90, |
| "ratio_in_domain": 0.250000, |
| "ratio_overall": 0.036058 |
| }, |
| "C": { |
| "name": "State Updating", |
| "count": 150, |
| "ratio_in_domain": 0.416667, |
| "ratio_overall": 0.060096 |
| }, |
| "D": { |
| "name": "State Abstraction", |
| "count": 59, |
| "ratio_in_domain": 0.163889, |
| "ratio_overall": 0.023638 |
| } |
| } |
| }, |
| "OPENWORLD_QA": { |
| "total_episodes": 30, |
| "episode_ratio": 0.144231, |
| "total_qa_pairs": 360, |
| "qa_ratio": 0.144231, |
| "problem_types": { |
| "A": { |
| "name": "Recall", |
| "count": 98, |
| "ratio_in_domain": 0.272222, |
| "ratio_overall": 0.039263 |
| }, |
| "B": { |
| "name": "Causal Inference", |
| "count": 95, |
| "ratio_in_domain": 0.263889, |
| "ratio_overall": 0.038062 |
| }, |
| "C": { |
| "name": "State Updating", |
| "count": 107, |
| "ratio_in_domain": 0.297222, |
| "ratio_overall": 0.042868 |
| }, |
| "D": { |
| "name": "State Abstraction", |
| "count": 60, |
| "ratio_in_domain": 0.166667, |
| "ratio_overall": 0.024038 |
| } |
| } |
| } |
| } |
| } |
|
|