This PR generalizes perf_test_multiple to include two additional important cases: when the true model and estimation model admit different numbers of parameters, and when the true model has a timestep update that must be considered. In the first case, both the estimated and true modelparameter vectors are now correctly recorded, while the loss is restricted to those parameters shared between the two (aligning at the right, in keeping with NumPy convention). In the second, a call to true_model.update_timestep is now made after each simulate_experiment call in the updater loop.
This PR generalizes
perf_test_multiple
to include two additional important cases: when the true model and estimation model admit different numbers of parameters, and when the true model has a timestep update that must be considered. In the first case, both the estimated and true modelparameter vectors are now correctly recorded, while the loss is restricted to those parameters shared between the two (aligning at the right, in keeping with NumPy convention). In the second, a call totrue_model.update_timestep
is now made after eachsimulate_experiment
call in the updater loop.