# Fashion-MNIST subset — S10 data bundle

`fashion_subset.npz` is a **6,000-image Fashion-MNIST subset**: 5,000 training images (500 per class) and 1,000 validation images (100 per class), 28×28 grayscale, 10 classes. It is 2.6 MB.

## Where it comes from

Built from Fashion-MNIST with `build_data.py` (prototype, seed 42) and bundled so the lab has **no runtime download**. Fashion-MNIST is MIT-licensed; the images are clothing silhouettes, with no people, faces or brands.

## Licence and reuse

Fashion-MNIST is MIT-licensed (Han Xiao, Kashif Rasul and Roland Vollgraf). The subset is redistributed for teaching; keep the attribution in the notebook and do not present it as the unit's real product photography. It is a proxy for the product-photo classifier, not a production dataset.

## Fields

- `train_images`, `train_labels`: 5,000 × 28 × 28 `uint8`, labels 0–9.
- `val_images`, `val_labels`: 1,000 × 28 × 28 `uint8`.
- `classes`: the ten class names.

## How to get it

- **Mac:** already at `data/fashion_subset.npz`.
- **Kaggle:** upload it to the notebook working directory, or place it under `/kaggle/input/` and set `COURSE_DATA`.
- **Faculty RTX / RunPod:** copy the same file to the working directory.
