# Amazon Computers

**Task**: Product category classification (co-purchase)  
**Size band**: medium · **Label type**: semantic  
**Label column**: `category` · **Converter**: `converters/convert_npz.py`

Amazon co-purchase graph of computer products; 10 category classes, 767 binary bag-of-words review features.

## Converted graphs (neext/)

| graph | nodes | edges | classes | feature cols | isolated | class counts |
|---|---|---|---|---|---|---|
| default | 13,752 | 245,861 | 10 | 767 | 281 | 4: 5,158, 8: 2,156, 1: 2,142, 2: 1,414, 7: 818, 3: 542, … |

*Conversion notes*: shchur gnn-benchmark npz; CSR adjacency symmetrized. 767 binary w_* feature columns.

## Source

- [amazon_electronics_computers.npz](https://raw.githubusercontent.com/shchur/gnn-benchmark/master/data/npz/amazon_electronics_computers.npz) — 31,921,488 bytes, sha256 `736ba1d9fd85eac2…`, fetched 2026-07-23

**License**: MIT (shchur/gnn-benchmark packaging)  
**Citation**: Shchur, Mumme, Bojchevski, Günnemann. Pitfalls of Graph Neural Network Evaluation. R2L @ NeurIPS 2018.

- https://github.com/shchur/gnn-benchmark

## Caveats

- Contains isolated nodes; largest-component filtering will drop some labeled nodes.

---
*Generated by converters/make_cards.py; stats from metadata.json. Raw files: `source/`. NEExT tables: `neext/`.*