Skip to content
/ DDUp Public

An implementation of the SIGMOD23 paper: Detect, Distill and Update: Detect, Distill and Update: Learned DB Systems Facing Out of Distribution Data

License

Notifications You must be signed in to change notification settings

meghdadk/DDUp

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DDUp; Detect, Distill and Update: Learned DB Systems Facing Out of Distribution Data

License: Apache

Introduction

The code includes instantiation of three learned database systems: DBEst++, Naru, and TVAE.

To develope DDUp, we have used the source codes published by the respective works at below:

DBEst++

Naru

TVAE

Setup

To increase reproducibility, we have created subdirectories for each dataset within each model. The latest codes for each case could be found inside the directory

For DBEst: The train/update procedures are located in the MDN.py. The evaluation procedures are located in benchmarking.py

For Naru: The train/update procedures are located in the incremental_train.py. The evaluation procedures are located in eval_model.py

For TVAE: The train/update procedures are located in the tvae_train.py. The evaluation procedures are located in benchmarking.py

The codes are tested for Python3.6 and Pytorch 1.9

Datasets

The experiments in the paper are for six public datasets. For DBest++, we have used a query template with two columns and have added the modified datasets in the related folders. For TVAE, we have used a samller (1m) sample of DMV dataset, as it was too expensive to train TVAE on the full data.

The link to some of the datasets:

Census

Forest

DMV

References

If you find this repository useful in your work, please cite our SIGMOD23 paper:

@article{kurmanji2023detect,
  title={Detect, Distill and Update: Learned DB Systems Facing Out of Distribution Data},
  author={Kurmanji, Meghdad and Triantafillou, Peter},
  journal={Proceedings of the ACM on Management of Data},
  volume={1},
  number={1},
  pages={1--27},
  year={2023},
  publisher={ACM New York, NY, USA}
}

About

An implementation of the SIGMOD23 paper: Detect, Distill and Update: Detect, Distill and Update: Learned DB Systems Facing Out of Distribution Data

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages