If your project contains data that has been newly created (i.e., which is not otherwise (publicly) available yet; including derived data sets), you are required to include a documentation of that data in your project.
Instances of “new data” may included, but are not restricted to be:
- data scraped from websites
- data gathered via APIs
- manually labeled data
- e.g., to assign GDP per capita to a list of countries
- e.g., to classify a music label as a major versus independent label
- data derived from secondary data (e.g., a cleaned data set; making explicit how you cleaned the data is important for future use of that data)
Think of “new data” as any data that feeds into one of the pipeline stages in your project; it really needs not to be “big” data, but can simply consist of a
.csv file with
names and associated labels (e.g., as in the case of countries –> GDP per capita).
Describe your raw data
Ideally, your data description includes the very elaborate questions outlined in
Datasheets for datasets by Gebru, Timnit, et al. (2018).
We strongly refer you to the original paper, which explains in detail the seven key ingredients of a proper dataset documentation. Below, we have reproduced these questions, and we recommend you to include those as a
readme.txt, together with your datasets. For derived data, it may be enough to point to a relevant source code file, and provide a list of variables and their operationalization.
That’s a lot of documentation. So - if you don’t have time, go with the bigger picture and answer the main questions only.
========================================================== D A T A S E T D E S C R I P T I O N ========================================================== Name of the dataset: ---------------------------------------------------------- 1. Motivation of data collection (why was the data collected?) [...] 2. Composition of dataset (what's in the data?) [...] 3. Collection process (how was the data collected?) [...] 4. Preprocessing/cleaning/labeling (how was the data cleaned, if at all?) [...] 5. Uses (how is the dataset intended to be used?) [...] 6. Distribution (how will the dataset be made available to others?) [...] 7. Maintenance (will the dataset be maintained? how? by whom?) [...]