Frontloading Visualization: A discovery driven approach to efficient analytics

  • BLOG
  • July 29th, 2019

Daniel Kohn, a Brooklyn-based painter and conceptual artist, is helping genetic scientists at the Albert Einstein College of Medicine in NY move out of their comfort zone. He is helping them answer unconventional questions like “…What if the data were turned sideways? Or, upside down? Or, what if you could click on a point on the plotted data and see another dimension…”

It is not just in the field of genetics. Organizations across the industry spectrum are looking at ways to interpret and represent data. In Mu Sigma, we believe that the science of data interpretation in combination with the art of visualization can make the consumption of analytics commonplace. And it is not just about interpreting data at the insight stage, rather at the point of analysis. This process helps us discover patterns that otherwise get lost in the quest to interpret the complete data.

Below are three modes of visualization that are being tried by the proponents of higher dimensionality:

Slice when it does not make sense: We can notice interesting patterns and variance emerge when an n-dimensional data is sliced into multiple 3-dimensional subsets. However, contrary to the traditional machine learning approach of using Principal Component Analysis or Multidimensional Scaling, Topological Data Analysis leverages clusters to project information. That is, the entire data is represented as a network with each point representing multiple points yielding similar information, and each connection highlighting the proportion of information common between the clusters. This reduction simplifies the comprehension and visibility.

Shadow is self: Here, we take a higher dimensional object and its shadow on a 2/3-dimensional canvas. The idea here is that some datasets do not lose their properties when forced to physical modifications like a stretch or shadow; for example, a spherical ball remains a circle no matter how it is projected on to a 2-D canvas. However, take a cuboid and the shadows represent different meaning at different projections. These shadows take different shape depending upon the variance at the edges, and thus help us understand the boundaries, reflecting what matters, and does not.

Change the point: Varying a points’ attributes such as time, color, shape can help us see if there something worth being a datum. These minute changes can transform the entire property of the entity and result in insights around the usefulness of that attribute. For example, in the image below, Buckminster Fuller idealized the color progression for a 4D tower. Changing the dimensionality of points at intersection reflect how the color changes from darkness to the yellow of dawn.