https://github.com/calgo-lab/error-paper

Code for the Paper "Towards Realistic Error Models for Tabular Data" submitted to the Journal of Data and Information Quality

https://github.com/calgo-lab/error-paper

Science Score: 26.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • CITATION.cff file
  • codemeta.json file
    Found codemeta.json file
  • .zenodo.json file
    Found .zenodo.json file
  • DOI references
  • Academic publication links
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (9.0%) to scientific vocabulary
Last synced: 9 months ago · JSON representation

Repository

Code for the Paper "Towards Realistic Error Models for Tabular Data" submitted to the Journal of Data and Information Quality

Basic Info
  • Host: GitHub
  • Owner: calgo-lab
  • Language: Jupyter Notebook
  • Default Branch: main
  • Homepage:
  • Size: 256 MB
Statistics
  • Stars: 1
  • Watchers: 3
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Created almost 2 years ago · Last pushed 10 months ago
Metadata Files
Readme

README.md

Towards Realistic Error Models for Tabular Data

The notebooks in this repository were used to execute the experimental evaluation of our paper "Towards Realistic Error Models for Tabular Data". Specifically,

1) the notebook dataset_generation.ipynb contains the procedure we followed to generate datasets corresponding to the error scenarios we describe in our paper. 2) The notebook dataset_analysis.ipynb contains our analysis of the HOSP dataset. 3) The notebook plots.ipynb contains the procedure we use to generate the figures in our publication. It reads experiment's results from the error_paper/measurements/ directory -- check the notebook's code for details.

Installation

We use poetry to manage dependencies. Simply run poetry install to install all dependencies.

Experiments

In our experiments, we examine data cleaning and downstream machine learning task impact using tab_err. - In the first part of the data cleaning experiments, we generate various erroneous versions of the HOSP dataset and clean them with HoloClean (benchmarks/hosp-impact). - We then proceed to generate various erroneous versions of datasets bridges, beers, restaurant and cars and correct them with algorithms baran&raha, holoclean and renuver (benchmarks/cleaning-impact). - In the downstream machine learning task impact, we look at how ML models behave given data with various errors (benchmarks/ml_downstream_experiments).

Check the documentation in benchmarks/README.md for instructions on how to replicate our measurements.

Profiling

We also looked at the memory and runtime of tab_err using various error models and dataset sizes. See the directory benchmarks/profiling for examples.

Owner

  • Name: Cognitive Algorithms Lab
  • Login: calgo-lab
  • Kind: organization
  • Location: Germany

GitHub Events

Total
  • Delete event: 2
  • Push event: 42
  • Pull request review comment event: 6
  • Pull request review event: 2
  • Pull request event: 5
  • Create event: 7
Last Year
  • Delete event: 2
  • Push event: 42
  • Pull request review comment event: 6
  • Pull request review event: 2
  • Pull request event: 5
  • Create event: 7

Dependencies

poetry.lock pypi
  • 121 dependencies
pyproject.toml pypi
  • error-generation *
  • jupyter ^1.0.0
  • matplotlib ^3.9.2
  • pandas ^2.2.2
  • pyarrow ^16.1.0
  • python >=3.9,<3.12
  • seaborn ^0.13.2
benchmarks/holoclean/Dockerfile docker
  • python 3.7 build
benchmarks/holoclean/docker-compose.yml docker
  • hc36 latest
  • postgres 11
benchmarks/holoclean/requirements.txt pypi
  • enum34 ==1.1.6
  • gensim ==3.7.1
  • numpy ==1.16.1
  • pandas ==0.24.1
  • psycopg2-binary ==2.7.7
  • pyitlib ==0.2.0
  • pytest-xdist ==1.26.1
  • python-Levenshtein ==0.12.0
  • scikit-learn ==0.20.0
  • scipy ==1.2.1
  • sqlalchemy ==1.2.17
  • torch ==1.0.1
  • tqdm ==4.31.0