aai-institute / aai-institute/pyDVL
Implement LAVA method
- Dominant language
- Python
- Stars
- 146
- Forks
- 10
- PR merge metrics
- No merged PRs in 30d
Description
Implement Just, Hoang Anh, Feiyang Kang, Jiachen T. Wang, Yi Zeng, Myeongseob Ko, Ming Jin, and Ruoxi Jia. [Lava: Data valuation without pre-specified learning algorithms.](https://arxiv.org/abs/2305.00054) arXiv preprint arXiv:2305.00054 (2023).
The paper's code can be found in [this repository](https://github.com/ruoxi-jia-group/LAVA).
To compute Optimal Transport we can use either [POT](https://pythonot.github.io/) and/or [GeomLoss](https://www.kernel-operations.io/geomloss/). The latter is Pytorch specific whereas the former isn't.
Contributor guide
Research direction
Read the linked LAVA paper and its GitHub repository to understand the algorithm. Examine pyDVL's existing data valuation modules for integration patterns. Determine whether to use POT or GeomLoss for Optimal Transport, considering the project's dependencies. The implementation will require designing a new module, writing tests, and documenting the method.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- ai-infra-agents, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 30/100