T-SNE: A Dimensionality Reduction Tool for Design Data Visualisation

Abstract

One can argue that data is the 'new oil'. Yet more important than the sheer quantity of data is the question, in the context of architecture and design, how data is represented in design, as this is becoming a more relevant question to the architecture profession. We argue that data, in particular n-dimensional, is often hidden even in BIM models. Hence we propose a new way of understanding the space by (1) generate and integrate space analytics data using space syntax method as well as space usage data and (2) visualise the data using t-Distributed Stochastic Neighbour Embedding (t-SNE), an unsupervised learning and dimensionality reduction tool to help intuitively display high dimensions of data. This approach may help to discover the 'hidden layers' of the building information that may be otherwise omitted. This investigation, its proposed hypothesis, methodology, implications, significance and evaluation are presented in the paper.