deepshopper
deepshopper-reward-pairs
DeepShopper Reward pairwise-preference data
(need, chosen=gold outfit, rejected=corrupted outfit, neg_type) pairs for Bradley-Terry
reward training. 93,020 train / 63,584 test, balanced over 5 corruption types:
gender_flip, item_swap, duplicate_role, count_drop, cross_need. Built (scripts/build_reward_pairs.py)
from the gold AMZ bundles via the frozen deepshopper-mapper-reward-splits.
Trains flavianv/qwen4b-reward-pairwise-v1. Code: https://github.com/clijo/reco-rl (branch… See the full description on the dataset page: https://huggingface.co/datasets/flavianv/deepshopper-reward-pairs.deepshopper-mapper-reward-splits
DeepShopper frozen splits (mapper / reward)
Deterministic, leakage-safe train/test splits used across DeepShopper. Split assignment is a
stable sha1(need) hash (same need never crosses train/test; reproducible). Gender-stratified.
Contains: fashionrec_task1_mapper/{train,test}, amz_mapper/{female,male,other}.{train,test}
(need→plan), and amz_reward_bundle/{female,male,other}.{train,test} (need→outfit, reward
positives). Generated by scripts/make_mapper_reward_splits.py. Code:… See the full description on the dataset page: https://huggingface.co/datasets/flavianv/deepshopper-mapper-reward-splits.deepshopper-reducer-sft
DeepShopper Reducer SFT
need + mapper query plan + per-role candidate pool → select the final outfit (by handle).
17,941 train rows (k=5 retrieved candidates per slot + ~30% distractors: ≥1 wrong-gender,
≥1 cross-role, rest off-style/age/popular), + held-out test. Candidates use ephemeral
per-row handles (prevents id memorization); target = soft-gold (closest retrievable to the
purchased item, gender-checked). Trains flavianv/qwen4b-reducer-v0. Code:… See the full description on the dataset page: https://huggingface.co/datasets/flavianv/deepshopper-reducer-sft.deepshopper-mapper-gender-transfer-mens