The Chicago crime tensor

One calendar year of Chicago crime reports, aggregated into an order-4 tensor of day-of-week × hour × community area × crime type — shape (7, 24, 78, 10), 131,040 cells.

Nothing is downloaded as a file: the tensor is built from one aggregated query that asks the city’s data portal to count for us, which is what a count tensor is. Only the ten commonest crime types are kept, so the type axis is short enough to read.

It is a hard tensor for squared error: 41% of the cells are zero — two in five.