DACE: A Light-Weight Learned Cardinality Estimation with Progressive Knowledge Distillation and Data Drift Adaptation

This repo provides the code of DACE. In this paper, we propose a framework that can be applied to various models to obtain lightweight CE. There are more details on the implementation of the experiments in the supplementary material.

Requirement

• OS: Ubuntu 18.04
• Packages:

Python 3.8.5
torch 1.7.1
Tensorflow 2.10
numpy, scipy, psycopg2, argparse

Run Pipeline

1. Progressive Knowledge Distillation on stable data.

python run_kd.py --backbone FACE --dataset JOB-light

2. Domain adapting and optimal transporting on dynamic data landscape.

python run_ds.py --backbone FACE --dataset JOB-light