KO
|
EN
gitlite โ search
Search
#javascript
#python
#hacktoberfest
#react
#ai
#typescript
#llm
#go
#golang
#android
#machine-learning
#rust
#deep-learning
#linux
D3M
โ 13
Open GitHub โ
Debiasing Through Data Attribution
Download README (.md)
Explore Similar Repositories
tabby-backend-py
:
Tabby (self-hosted AI coding assistant) server in 20 lines of python
Perdoname
:
๐ Proyecto para los bros que necesitan el perdรณn
Token-bridge
:
A token bridge utilizing PortalBridge to facilitate the seamless transfer of assets across multiple blockchain networks.
sdcore-helm-charts
:
Helm charts used for SD-Core packaging
survey-scanner
:
This repository publishes the sources of Internet-Wide Scans observed by NICTER darknet monitoring.
// repository documentation
Was this content helpful?
โ 0
(0 ratings)
Select Rating:
โ
โ
โ
โ
โ
Submit Feedback
Recent Feedback
×
Download README
Do you want to download the
README.md
file for
D3M
?
Download (.md)
# Data Debiasing with Datamodels (D3M): Improving Group Robustness via Dataset Selection To run our code, first install our library by cloning this repository and running: ``` cd d3m pip install -e . ``` Then, you can use the following code to debias your dataset: ``` from d3m import D3M # Any pytorch model model = ... # A list of model checkpoints, i.e. a list of "state_dict" objects checkpoints = ... # Pytorch dataloaders for the training and validation sets train_dataloader = ... val_dataloader = ... # A list of group indices, i.e. a list of indices, where each element is # the index of the group of the corresponding validation example group_indices = ... d3m = D3M(model, checkpoints, train_dataloader, val_dataloader, group_indices) debiased_train_indices = d3m.debias() # Use debiased_train_indices to create a new dataloader debiased_train_dataloader = ... # Train the model with the new dataloader train(model, debiased_train_dataloader, ...) ... ```
