---
pretty_name: mind-data
license: mit
viewer: false
configs:
- config_name: Assist17
  data_files: "Assist17/response.csv"
  default: true
- config_name: FrcSub
  data_files: "FrcSub/response.csv"
- config_name: Math1
  data_files: "Math1/response.csv"
- config_name: Junyi734
  data_files: "Junyi734/response.csv"
- config_name: Math2
  data_files: "Math2/response.csv"
- config_name: Matmat
  data_files: "Matmat/response.csv"
- config_name: NeurIPS20
  data_files: "NeurIPS20/response.csv"
- config_name: XES3G5M
  data_files: "XES3G5M/response.csv"
---

# mind-data

Raw cognitive diagnosis datasets prepared for `mind-toolkit`.

Each supported dataset is exposed as an individual Hugging Face configuration (subset). The Hub Dataset Viewer is intentionally disabled because these files are raw response logs and auxiliary matrices rather than human-friendly preview tables.

## Structure

Each dataset is stored in its own folder. Core files use a shared naming convention:

- `response.csv`: raw response triplets `(entity_id, exercise_id, label)`
- `q_matrix.csv`: exercise-knowledge incidence matrix
- `metadata.yaml`: dataset statistics and auxiliary file manifest

Some datasets also include extra structured files when the source provides them, for example:

- `semantic_relation.csv`
- `concept_hierarchy.csv`

These files are retained only when they contain reusable task information beyond the basic response log and Q-matrix.

## Included datasets

- Assist17
- FrcSub
- Math1
- Junyi734
- Math2
- Matmat
- NeurIPS20
- XES3G5M

## Notes

This repository stores raw, unsplit data. Train/validation/test splitting is handled by `mind-toolkit` at experiment time.