07 Sep Data Visualisation as Art
When we are working with data all day every day, one of the things that often strikes me is the inherent beauty in the visualisations that we create. Here’s one from the other day, where we are using t-SNE to better understand the inherent structure in some data. Playing around with the number of iterations and perplexity settings allows us to get different intermediate (i.e. non-converged) visualisations.
From Machine Learning to Art
Our motivation for visualisation is as part of the toolkit for improving business performance through AI/machine learning. People like Marcin Ignac from VariableIO are taking this idea to another level. I met Marcin at a recent event, and he’s coming at it from a completely different and interesting perspective. He is using his background in Computer Science to produce data-driven art, and his tools of the trade are code and IDEs, not physical media. You might have seen some VariableIO’s work on the tube in London:
VariableIO images at a tube station
You can see the other TfL designs in more detail here. Actually from a data science perspective, this is arguably the less interesting (but visually arresting!) work from Marcin & co, as the penguin or tea cup for TfL are not data-driven in any way per se. They have used 3D models of everyday objects or animals) and then used TfL colours as an inspiration for their generative system.
However there’s lots of directly data-driven examples on the VariableIO site so take a look, e.g. this visualisation of twitter traffic related to a museum exhibit. To generate this kind of work, they use their own open source WebGL code called PEX.
Make sure you also check this Every day of my life work out – it’s a visualisation of Marcin’s life using his computer usage statistics from the last 2.5 years.
Finally, getting back to where I started on t-SNE, one other thing to throw in – this t-SNE mapping from Google is definitely worth checking out also. They’ve used t-SNE to arrange a map of artworks based on their visual similarity.
Thanks to VariableIO for permission to use the images and videos in this post.