README.md

---
pretty_name: DataIntern Data Spaces
license: other
language:
- en
size_categories:
- 10K<n<100K
tags:
- data-agents
- benchmark
- duckdb
configs:
- config_name: tables
  data_files: catalog/tables.parquet
- config_name: files
  data_files: catalog/files.parquet
- config_name: documents
  data_files: catalog/documents.parquet
---

# DataIntern data spaces

The fifteen data spaces of DataIntern, a benchmark of whether data agents learn from the queries they have completed.
Each space holds DuckDB tables, the raw and curated files behind them, and reference documents, all from public
sources (450 tables and views, 5,678 data files, 13,262 documents, 67 GB). Queries and grading are released with the code.

```
catalog/                  tables, files and documents of every space
spaces/<space>/
  duckdb/<space>.duckdb   views use paths relative to spaces/<space>/
  datasets/               raw and curated files
  documents/              reference documents
  schema/                 ER diagram, join map, table inventory
  workspace/              docs_schema/, modeling.yaml, sources.yaml
```

```bash
hf download DataInternSubmission/DataIntern-DataSpaces --repo-type dataset --include "spaces/nyc_urban_mobility/*"
```