Google open-sourced Facets: a data visualisation tool to explore data for machine learning scientists.
Facets aim is to make big data sets understandable and interpretable. Right now it's difficult to understand a data set intuitively, and it takes developers a lot of time to create graphs for this purpose. Facets can help developers find nuances and insights in large data sets.
Facets consists of two visualisations: Facets Overview and Facets Dive. Overview takes input feature data and analyses it feature by feature. Its goal is to give developers a quick understanding of the features in their dataset, their distribution, and unexpected values. Facets Dive is an interactive tool in which developers can bucket items in multiple dimensions based on their feature value.
The release is part of the PAIR initiative. A research group devoted to advancing research and design of people-centric AI systems. Their goal is to research, invent technology, and create better tools to make the interaction between people and artificial intelligence more productive, enjoyable, and fair. With their tools engineers will be able to better understand machine learning systems, so everyone can benefit from breakthroughs in artificial intelligence.
Developers need to clone the Facets Git repository to install Facets. After installing an extension to Jupyter, developers can also add visualisations to their Jupyter Notebooks by installing it as an extension. At the moment the visualisations only work in the Chrome browser, but this will be resolved in the future.
Facets is an addition to the tool Tensorflow users already often use: Tensorboard. Facets can also analyse the performance of a model. It is not sure if Facets wants to compete with Microsofts visualisation tool SandDance, or if it will become an addition to the toolkit of big data analysts.
Developers can find more information about Facets on their demo website or their Github page.