Skip to content

dennisschneider-ml/datasets

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

25 Commits
 
 
 
 
 
 
 
 

Repository files navigation

datasets

This is a template project which adds a reproducible way to preprocess datasets.

Installation

Clone this repository and run sh install.sh.

Why would I want to use datasets?

If you use datasets in a project, you probably want other researchers to be able to easily reproduce your results, beginning with using the same datasets. Most research projects provide a step-by-step instruction on how to get these datasets. But why not encapsule all this into a single command?
judo-datasets is an extension of the judo project and provides an easy and reproducible way to generate datasets by simply invoking the make command, generating only these datasets which have not been processed yet, which also makes it suitable for using while continuously adding new datasets to the project.

Great, so how does it work?

First, run make once to create the ./data/ directory in the root of your project.

After finding a dataset you want to use, add an empty directory in this ./data/ directory. Run make again to create the recommended directory structure for the new dataset.

This results in the following data tree:

 data
└──  dataset
   ├──  get_original_data.sh
   ├──  original
   ├──  preprocess.py
   └──  preprocessed

At this point you can place the commands for populating the original-subdirectory in the get_original_data.sh file. Furthermore, the preprocess.py script enables us to preprocess the original datasets for further usage. By invoking make once more, the sh script is invoked and the files are downloaded (or however you choose to retrieve these files). And, since no files have been preprocessed yet, all files in original are then suitably converted into preprocessed files by running them through preprocess.py. It conveniently already employs multithreading and splits up the dataset among all CPU-cores. You simply have to implement the preprocessing-logic in the process-function in the python-script.

The Makefile conveniently detects which datasets need a file-structure, which files have to be downloaded and which have unprocessed data. So mass-processing multiple datasets over night is as easy as it gets.
Furthermore, by adding the original and preprocessed subdirectories in the gitignore, a researcher can reproduce all datasets by simply running make.

About

A Plugin for the judo project, enabling a reproducible way of dataset-management.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published