MolecularDatasetCurationGuide / sections /04_create_dataset_card.md
woodsh's picture
Fix small typo in sections/04_create_dataset_card.md
55d3ecc verified
## **4 Add an informative README.md**
The `README.md` is a markdown file that is displayed when goes to the front page for the dataset. It should give appropriate context for the dataset and guidance on how to use it. As a template, consider having these sections--where the parts in brackets should be filled in. See the [MIP](https://huggingface.co/datasets/RosettaCommons/MIP/blob/main/README.md) dataset as an example.
| \# \<DATASET TITLE\>
\<short descriptive abstract the dataset\>
\#\# Quickstart Usage
\#\#\# Install HuggingFace Datasets package
Each subset can be loaded into python using the HuggingFace \[datasets\](https://huggingface.co/docs/datasets/index) library.
First, from the command line install the \`datasets\` library
$ pip install datasets
Optionally set the cache directory, e.g.
$ HF\_HOME=${HOME}/.cache/huggingface/
$ export HF\_HOME
then, from within python load the datasets library
\>\>\> import datasets
\#\#\# Load model datasets
To load one of the \<DATASET ID\> model datasets, use \`datasets.load\_dataset(...)\`:
\>\>\> dataset\_tag \= "\<DATASET TAG\>"
\>\>\> dataset \= datasets.load\_dataset(
path \= "\<HF PATH TO DATASET\>",
name \= f"{dataset\_tag}",
data\_dir \= f"{dataset\_tag}")\['train'\]
and the dataset is loaded as a \`datasets.arrow\_dataset.Dataset\`
\>\>\> dataset
\<RESULT OF LOADING DATASET MODEL\>
which is a column oriented format that can be accessed directly, converted in to a \`pandas.DataFrame\`, or \`parquet\` format, e.g.
\>\>\> dataset.data.column('\<COLUMN NAME IN DATASET\>')
\>\>\> dataset.to\_pandas()
\>\>\> dataset.to\_parquet("dataset.parquet")
\#\#\# \<BREIF EXAMPLE OF HOW TO USE DIFFERENT PARTS OF THE DATASET\>
\#\# Dataset Details
\#\#\# Dataset Description
\<DETAILED DESCRIPTION OF DATASET\>
\- \*\*Acknowledgements:\*\*
\<ACKNOWLEDGEMENTS\>
\- \*\*License:\*\* \<LICENSE\>
\#\#\# Dataset Sources
\- \*\*Repository:\*\* \<URL FOR SOURCE OF DATA\>
\- \*\*Paper:\*\* \<APA CITATION REFERENCE FOR SOURCE DATA\>
\- \*\*Zenodo Repository:\*\* \<ZENODO LINK IF RELEVANT\>
\#\# Uses
\<DESCRIPTION OF INTENDED USE OF DATASET\>
\#\#\# Out-of-Scope Use
\<DESCRIPTION OF OUT OF SCOPE USES OF DATASET\>
\#\#\# Source Data
\<DESCRIPTION OF SOURCE DATA\>
\#\# Citation
\<BIBTEX REFERENCE FOR DATASET\>
\#\# Dataset Card Authors
\<NAME/INFO OF DATASET AUTHORS\> |
| :---- |
## **5 Add Metadata to the Dataset Card**
### **Overview**
A the top of the \`README.md file include metadata about the dataset in yaml format
\---
language: …
license: …
size\_categories: …
pretty\_name: '...'
tags: …
dataset\_summary: …
dataset\_description: …
acknowledgements: …
repo: …
citation\_bibtex: …
citation\_apa: …
\---
For the full spec, see the Dataset Card specification
* [Dataset Card Documentation](https://huggingface.co/docs/hub/en/datasets-cards)
* [Dataset Card Specification](https://github.com/huggingface/hub-docs/blob/main/datasetcard.md?plain=1)
* [Dataset Card Template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/datasetcard_template.md)
To allow the datasets to be loaded automatically through the datasets python library, additional info needs to be in the header of the README.md. It should reflect how the [repository is structured](https://huggingface.co/docs/datasets/en/repository_structure)
configs:
dataset\_info:
While it is possible to create these by hand, it highly recommended allowing it to be created automatically when uploaded via loading the dataset locally with [datasets.load\_dataset(...)](https://huggingface.co/docs/datasets/en/loading), then pushing it to the hub with [datasets.push\_to\_hub(...)](https://huggingface.co/docs/datasets/main/en/package_reference/main_classes#datasets.DatasetDict.push_to_hub)
* [Example of uploading data using push\_to\_hub()](https://huggingface.co/datasets/RosettaCommons/MegaScale/blob/main/src/03.1_upload_data.py)
* See below for more details about how to use push\_to\_hub(...) for different common formats
### **Metadata fields**
#### License
* If the dataset is licensed under an existing standard license, then use it
* If it is unclear, then the authors need to be contacted for clarification
* Licensing it under the Rosetta License
* Add the following to the dataset card:
license: other
license\_name: rosetta-license-1.0
license\_link: LICENSE.md
* Upload the Rosetta [LICENSE.md](https://github.com/RosettaCommons/rosetta/blob/main/LICENSE.md) to the Dataset
#### Citation
* If the dataset has a DOI (e.g. associated with a published paper), use [doi2bib.org](http://doi2bib.org)
* [DOI → APA converter](https://paperpile.com/t/doi-to-apa-converter/):
#### tags
* Standard tags for searching for HuggingFace datasets
* typically:
\- biology
\- chemistry
#### repo
* Github, repository, figshare, etc. URL for data or project
#### citation\_bibtex
* Citation in bibtex format
* You can use https://www.doi2bib.org/
#### citation\_apa
* Citation in APA format