huggingface / huggingface/datasets
Add a `lazy_map` method to `Dataset` and `DatasetDict`
Open
enhancement
- Dominant language
- Python
- Stars
- 22k
- Forks
- 3.4k
- Avg merge
- 5d 7h
- Merged PRs (30d)
- 17
Description
The library is great, but it would be even more awesome with a `lazy_map` method implemented on `Dataset` and `DatasetDict`. This would apply a function on a give item but when the item is requested. Two use cases:
1. load image on the fly
2. apply a random function and get different outputs at each epoch (like data augmentation or randomly masking a part of a sentence for BERT-like objectives).
Contributor guide
Assessment
This issue has not been assessed yet.