A Plaque Test for Redundancies in Relational Data - Reproduction Package

This is a reproduction package for the paper "A Plaque Test for Redundancies in Relational Data" by Christoph Köhnen, Stefan Klessinger, Jens Zumbrägel and Stefanie Scherzinger, published in the QDB workshop co-located with VLDB 2023. It contains a program code as a submodule and experiments on real-world datasets, executed with this program.

This reproduction package provides a Docker container.

The datasets are located in the directory experiments/tables, the original results in the directory experiments/results.

Setting up the reproduction package

Clone the reproduction package and the program code using the parameter --recurse-submodules:

git clone https://github.com/sdbs-uni-p/plaque_test.git --recurse-submodules

You can set up the package on the host system or on a docker container.

Setup on host system

Enter the submodule, set up the program and copy the jar file into the folder experiments:

cd relational_information_content
mvn clean install
cd ..
cp relational_information_content/target/relational_information_content-1.0-SNAPSHOT-jar-with-dependencies.jar experiments/relational_information_content.jar

Setting up a docker container

docker build -t plaque_test_experiments .
docker run -it plaque_test_experiments

Running the experiments

On the host system or in the docker container, enter the experiments folder and run the experiments:

cd experiments
./run_experiments.sh

Optionally, a description can be added:

• --description <description>: A description to appear in the name of the results folder. The default is the empty string.

The experiments can take several hours. To reduce the runtime, use the provided options:

• --threads <number>: Multiple experiments are parallelized in <number> threads. The default value is 1.
• --timeout <seconds>: Each entropy computation aborts after <seconds> seconds. The default value is 86400 seconds, i.e., 24 hours.
• --skip-entropy-calc <boolean>: Only produces charts using the results in the folder results if <boolean> is true. The default value is false.

The results of the experiments are saved in the folder results_<description>_threads_<number>_timeout_<seconds>s or results_<description>_skip_entropy_calculations.

Folder structure

The repository consists of the folders relational_information_content and experiments.

The folder relational_information_content

This folder contains the relational information content program code as a submodule.

To run the experiments on a specified version, enter this folder and checkout this version.

The folder experiments

This folder consists of the following components:

• Python scripts to run the relational information content program for experiments and to create charts.
• The script run_experiments.sh to run all the experiments and create charts.
• The folder tables containing the data and configuration files.
• The folder results with the original experiment results. It contains:
  • The files accuracy.pdf and accuracy_compact.pdf which plot the sufficient number of iterations for the Monte Carlo method to achieve an accuracy with a specified confidence.
  • The folders adult, echocardiogram, iris and ncvoter. Each folder contains a CSV file with the entropy values computed with the Monte Carlo method (100,000 iterations) on an excerpt consisting of the first 150 rows (or all 132 rows for echocardioram), a CSV file with the runtime of the computation and a PDF file which illustrates the plaque.
  • The folder satellites. The dataset is considered on the first 150 rows only. It contains the following files:
    • satellites.pdf: Illustrates the plaque based on the entropy values computed with the Monte Carlo method (100,000 iterations).
    • satellites_monte-carlo-100-000_rows_150.csv: Contains the entropy values computed with the Monte Carlo method (100,000 iterations).
    • satellites_monte-carlo-100-000_rows_150_r109-128.csv: Contains the entropy values from satellites_monte-carlo-100-000_rows_150.csv, but restricted to the rows from 109 to 128.
    • satellites_runtimes.csv: Contains the runtimes of the Monte Carlo computations for different numbers of rows (e.g., excerpt containing first 10 rows) and iterations.
    • satellites_opt-off_runtimes.csv: Contains the runtimes of the unoptimized computations for different numbers of rows.
    • satellites_opt-on_runtimes.csv: Contains the runtimes of the optimized computations (identify entropy values of 1 directly and reduce the table to its relevant rows and columns) for different numbers of rows.
    • satellites_runtimes.pdf and satellites_runtimes_compact.pdf: Illustrate the runtimes from satellites_runtimes.csv in a heatmap.
    • satellites_zoom.pdf: Illustrates the entropies in an excerpt of the satellite dataset.
    • satellites_with_zoom.jpg: Shows the plaque computed with the Monte Carlo method (100,000 iterations) zooming into the aforementioned excerpt.
    • satellites_hist.pdf: Contains a histogram about the frequency of entropy values in the given dataset.
    • satellites-monte-carlo-1-000_rows_150.csv and satellites-monte-carlo-1-000-000_rows_150.csv: Contain the entropy values computed with the Monte Carlo method for 1,000 resp. 1,000,000 iterations.
    • satellites-mc-1000.pdf and satellites-mc-1000000.pdf: Illustrate the plaque based on the entropy values in satellites-monte-carlo-1-000_rows_150.csv resp. satellites-monte-carlo-1-000-000_rows_150.csv.
    • satellites_entropy_diffs_hist.pdf: Contains a histogram about the frequency of differences between the entropies computed in satellites_monte-carlo-1-000_rows_150.csv and satellites_monte-carlo-1-000-000_rows_150.csv.

After running the experiments, the output folder contains the following files and folders (note that each visualization is only created if the necessary entropy computations are not aborted by a timeout):

• The files accuracy.pdf and accuracy_compact.pdf which plot the sufficient number of iterations for the Monte Carlo method to achieve an accuracy with a specified confidence.
• The folders adult, echocardiogram, iris and ncvoter. Each folder contains a CSV file with the entropy values computed with the Monte Carlo method (100,000 iterations) on an excerpt consisting of the first 150 rows (or all 132 rows for echocardioram), a CSV file with the runtime of the computation and a PDF file which illustrates the plaque.
• The folder satellites. The dataset is considered on the first 150 rows only. It contains the following files:
  • satellites_heatmap_entropies.pdf: Illustrates the plaque based on the entropy values computed with the Monte Carlo method (100,000 iterations).
  • satellites_heatmap_entropies_mc-1000.pdf: Illustrates the plaque based on the entropy values computed with the Monte Carlo method (1,000 iterations).
  • satellites_heatmap_entropies_mc-1000000.pdf: Illustrates the plaque based on the entropy values computed with the Monte Carlo method (1,000,000 iterations).
  • satellites_monte-carlo-<iterations>_rows_<num_rows>.csv: Contains the entropy values computed with the Monte Carlo method (<iterations> iterations), but only on a considered excerpt consisting of the first <num_rows> rows.
  • satellites_opt-off_optimizations-off_rows_<num_rows>.csv: Contains the entropy values computed without any optimizations, but only on a considered excerpt consisting of the first <num_rows> rows.
  • satellites_opt-on_optimizations-on_rows_<num_rows>.csv: Contains the entropy values computed with optimizations (identify entropy values of 1 directly and reduce the table to its relevant rows and columns), but only on a considered excerpt consisting of the first <num_rows> rows.
  • satellites_runtimes.csv: Contains the runtimes of the Monte Carlo computations for different numbers of rows and iterations.
  • satellites_opt-off_runtimes.csv: Contains the runtimes of the unoptimized computations for different numbers of rows.
  • satellites_opt-on_runtimes.csv: Contains the runtimes of the optimized computations for different numbers of rows.
  • satellites_runtimes.pdf and satellites_runtimes_compact.pdf: Illustrate the runtimes from satellites_runtimes.csv in a heatmap.
  • satellites_zoom.pdf: Illustrates the entropies in an excerpt of the satellite dataset.
  • satellites_hist.pdf: Contains a histogram about the frequency of entropy values in the given dataset.
  • satellites_entropy_diffs_hist.pdf: Contains a histogram about the frequency of differences between the entropies computed in satellites_monte-carlo-1-000_rows_150.csv and satellites_monte-carlo-1-000-000_rows_150.csv.

Datasets

The satellites dataset is retrieved from the input data of the Metanome tool. The datasets adult, echocardiogram, iris, and ncvoter are retrieved from HPI Potsdam.

About

To reference this work, please use the following BibTeX entry.

@inproceedings{DBLP:conf/vldb/KohnenKZS23,
  author       = {Christoph K{\"{o}}hnen and
                  Stefan Klessinger and
                  Jens Zumbr{\"{a}}gel and
                  Stefanie Scherzinger},
  title        = {A Plaque Test for Redundancies in Relational Data},
  booktitle    = {Joint Proceedings of Workshops at the 49th International Conference
                  on Very Large Data Bases {(VLDB} 2023), Vancouver, Canada, August
                  28 - September 1, 2023},
  series       = {{CEUR} Workshop Proceedings},
  volume       = {3462},
  year         = {2023}
}

All artifacts, including source code, data, and scripts, are available on Zenodo, DOI: 10.5281/zenodo.8220684.

The reproduction package was created by Christoph Köhnen. The experimental scripts were created by Stefan Klessinger and Christoph Köhnen.

This work was partly funded by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) grant #385808805.