Datasetpublic

rayna-rules/flippd-verify

Flippd benchmark — public verification subset Anonymized subset for independently reproducing the Flippd leave-X-out recommendation benchmark without model weights or the full dataset. metadata.jsonl — one row per listing: img_id, seller_id (anonymized), gender, category, color0, brand, price. embeddings.npz — precomputed vectors (model outputs, not weights) for every encoder scored in the study: resnet, clip, fc_nohead, fc_head (the study's private encoder), aligned to ids.… See the full description on the dataset page: https://huggingface.co/datasets/rayna-rules/flippd-verify.

sourceHugging Facemitupdated 27d agoView on Hugging Face

0likes82downloads

Dataset Card

Flippd benchmark — public verification subset

Anonymized subset for independently reproducing the Flippd leave-X-out recommendation benchmark without model weights or the full dataset.

—metadata.jsonl — one row per listing: img_id, seller_id (anonymized), gender, category, color0, brand, price.
—embeddings.npz — precomputed vectors (model outputs, not weights) for every encoder scored in the study: resnet, clip, fc_nohead, fc_head (the study's private encoder), aligned to ids.
—images/<img_id>.jpg — the primary photo of each listing.

Reproduce the statistics + plot with the verification repo: <https://github.com/RaynaArora/flippd-benchmark> — python verify.py.