Skip Navigation

Scout Archives

Home Projects Publications Archives About Sign Up or Log In

Facets: Know Your Data

"Better data leads to better models" is the motto of Facets, a toolkit for "understanding and analyzing machine learning datasets." Facets offers two powerful data visualization tools. The first, Overview, gives users a quick visual analysis of the distribution of values across the features of one or more datasets. Overview provides users with summary statistics that give the general shape of each feature of their dataset and may help identify issues like unexpected values, missing values for a large number of observations, training/serving skew, and train/test/validation skew. The second tool, Dive, is an interactive interface for exploring large numbers of data points at once. Dive visualizes the relationships between data points across all different features of a dataset and allows a data point to be bucketed in multiple dimensions. The tool can help users identify classifier failure, systematic errors, and potential new signals for ranking. Facets is an open-source software toolkit developed by the People + AI Research (PAIR) team at Google Research.
Archived Scout Publication URL
Scout Publication
Classification
GEM Subject
Language
Date of Scout Publication
July 9th, 2021
Date Of Record Creation
June 22nd, 2021 at 2:39pm
Date Of Record Release
June 23rd, 2021 at 8:26am
Resource URL Clicks
215
Add Comment

Comments

(no comments available yet)