Add H&M fashion recommendation dataset #2708
Conversation
for more information, see https://pre-commit.ci
ludwig/datasets/loaders/utils.py
Outdated
| return neg_items.tolist(), max(0, samples_required - available_samples) | ||
|
|
||
|
|
||
| def negative_sample( |
There was a problem hiding this comment.
I think we're exceeding the scope of the datasets API by performing negative sampling here. The datasets API intent is to return standard benckmark datasets 'as-is' (or as close to the original as we can get).
IMO the sampling implementation belongs in ludwig.data, and should be performed as a separate phase after the dataset is loaded (also sampling has hyperparameters neg_pos_ratio and log_pct).
Is this possible, or would it be too inefficient to merge article and customer features prior to negative sampling?
There was a problem hiding this comment.
I agree - removed the negative sampling from the loader and moved it to ludwig.data in this PR
Unit Test Results 6 files ± 0 6 suites ±0 3h 29m 52s ⏱️ - 4m 11s For more details on these failures, see this check. Results for commit 3f2ab35. ± Comparison against base commit 3504538. ♻️ This comment has been updated with latest results. |
|
I extracted the negative sampling into a separate PR |
No description provided.