Datasets:
| language: en | |
| license: apache-2.0 | |
| task_categories: | |
| - text-generation | |
| tags: | |
| - code | |
| - pandas | |
| - data-analysis | |
| size_category: small | |
| configs: | |
| - config_name: default | |
| data_files: | |
| - split: train | |
| path: train.jsonl | |
| - split: eval | |
| path: eval.jsonl | |
| # analyst-data | |
| Training corpus for analyst-v1: a small pandas-codegen model. | |
| - 613 train / 165 eval-pool chat-format pairs (eval pool = pairs with constructible test DataFrames, held out of training) | |
| - Sources: DS-1000 seed (real Kaggle-style data problems) + synthetic pairs generated from live Kaggle datasets (every pair execution-verified against the real DataFrame) | |
| Format: {"system": ..., "user": "Schema:... Question:...", "assistant": "<pandas code>"} | |