Skip to content

Comparison with standard methods

Each stage is to be run against an established alternative, with the same inputs, targets, hold-out data and measures, and with no variant fitted to traffic counts:

  • Population synthesis: the conditional variational autoencoder against iterative proportional fitting with a weighted draw, iterative proportional updating (PopulationSim) and weighted sampling of the training households.
  • Land use: against the original SILO Maryland model, calibrated with the same procedure.
  • Activities: MAST against the original activity set-up and a standard tour-based model.
  • Assignment: the same network and assignment with the original activity demand.

The comparison is updated after the final runs. The validation of each stage against observed data is on the validation pages.