Same models, independent simulations

I am looking to benchmark three models by running them multiple times and averaging their results.
The requirement is that each of the runs uses the same seed on the input and that the runs are independent from each other.
To this end I used the following logic:

for i in range( num_runs ):
    learned_model_mpes = LearningModel( ..., seed=seed + i )
    control_model_pes = LearningModel( ..., seed=seed + i )
    control_model_nef = LearningModel( ..., seed=seed + i )

    with nengo_dl.Simulator( learned_model_mpes ) as sim_mpes: sim_time )
    with nengo_dl.Simulator( control_model_pes ) as sim_pes: sim_time )
    with nengo_dl.Simulator( control_model_nef ) as sim_nef: sim_time )

with this snippet at the beginning of each LearningModel:
nengo_dl.configure_settings( stateful=False )

Is this overkill to ensure independence between the runs?

If you wanted to ensure that each run was independent, you’ll need to make a separate simulator object for each model, which is what you are doing in your code, so it looks good to me! :smiley:

And are the simulations independent between runs i.e., is a new Simulator created at each iteration of the for?

That should be the case, yes.