/ShapleyForDbFacts

Primary LanguageJupyter Notebook

Computing the Shapley Value of Tuples in Query Answering

This repository is the official implementation of "Computing the Shapley Value of Tuples in Query Answering".

Requirements

Prerequisites
  1. Installation of PostgreSQL >= 9.5
  2. Installation of ProvSQL
  3. Installation of c2d
Datasets

Download and import the following databases to PostgreSQL.

  1. IMDB
  2. TPC-H

Each table should be tracked by ProvSQL.

SET search_path TO public, provsql;

SELECT add_provenance('[Table name]');
SELECT create_provenance_mapping('[Table name]_id', '[Table name]', '[Table primary index]');

Example

Example.ipynb depict how to compute the exact Shapley value of tuples given a provenance as a d-DNNF.

Main results

notebooks directory depicts subset of the results presented in the paper.

Results include:

  • Exact computation present the computation of exact Shapley values. We investigate the running time as a function of the provenance complexity.
  • Inexact computation compare several approximation and heuristics over a wide range of metics.