Average Tripledecker

2020

We’ve been thinking a lot about this common local three-flat housing type in our neighborhood and we are interested in using machine learning to understand and uncover the DNA of everyday architecture. Triple-deckers have pretty standard plans but a huge range of elevation features. We were curious about whether there is an “average triple-decker.” In our spare time we tried to train a machine learning model to generate the Platonic ideal of a triple decker.

This research, which included building a novel data set of Google Streetview images of Boston's housing stock, helped us assess the essential elements of a Boston Tripledecker housing type and model a set of "average" tripledeckers. This work supports projects such as PLAN: Mattapan and other housing related design work.