Uber Technologies, Inc.
Estimating an optimal ordering for data compression

Abstract:

A system orders a set of objects to be optimally compressed using techniques such as delta encoding which compress data by saving differences between sequential objects. In one embodiment, the system determines transition costs associated with transitioning from each object in the set to each other object in the set. The system determines an ordering of the objects that minimizes a total of all the transition costs and compresses the set in the determined order. In another embodiment, the system determines an estimated optimal ordering for data compression. The system extracts feature values from the objects and applies a projection function to the feature values to generate scalar representations of the objects, for example, by projecting the feature values onto a space filling curve. An ordering of the scalar values for the objects determines an ordering for compressing the objects.