To generate (or 'hallucinate') neighborhoods based on the distribution of neighborhoods, and the roads and buildings within them.

• Because it is very interesting to see what a deep neural net can learn as a representation of some kind of high-dimensional data distribution
• Also: generating semi-real world geospatial information objects could help spatial planners to be inspired by auto-generated neighbourhood layouts.

• Get the neighborhoods data
• Create a vector representation of neighborhoods to use in a machine learning model
• Create a train/test split for the neighborhoods
• Persist the training and test data