Running Experiments

To collect data with simulations, we need to run an experiment. To create an experiment, one needs a system and a Collector (object that collects data while the simulation is running). Let's demonstrate how to run an experiment in Mavi:

Given we have a function create_system (which creates a system), the first thing to do is to create the experiment configurations

using Mavi.Experiments

system = create_system()

experiment_cfg = ExperimentCfg(
    tf=10, # Final simulation time
    root="experiment_data", # Where the experiment data will be saved
)

next, we need to create the collector configurations, it's possible to use existing ones or create a custom collector, let's use the DelayedCol, which always have the system state some time in the past (this collector is useful to have the state system just before something bad happens, allowing the coder to investigate what is happening)

using Mavi.Experiments

system = create_system()

experiment_cfg = ExperimentCfg(
    tf=10, # Final simulation time
    root="experiment_data", # Where the experiment data will be saved
)

collector_cfg = DelayedCfg(
    delay_time=4, # How far in the past the state is
)

now we can create an experiment and run it

using Mavi.Experiments

system = create_system()

experiment_cfg = ExperimentCfg(
    tf=10, # Final simulation time
    root="experiment_data", # Where the experiment data will be saved
)

collector_cfg = DelayedCfg(
    delay_time=4, # How far in the past the state is
)

experiment = Experiment(experiment_cfg, collector_cfg, system)

run_experiment(experiment)

all data collected by the Collector and other things will be inside in the root path, specified when creating ExperimentCfg. Here's a list (not complete) of what is saved in an experiment:

• data: data collected by the Collector.
• final_state: system state just after the experiment finished.
• configs.json: system configurations used.
• experiment_configs.json: experiment configurations used.

Using checkpoints

If your experiment takes a lot of time, it's a good ideia to run the experiment with checkpoints, in that way, an experiment can be resumed from a checkpoint if something bad happens. To do that, simply pass a CheckpointCfg to ExperimentCfg

experiment_cfg = ExperimentCfg(
    tf=10,
    root="experiment_data",
    checkpoint_cfg=CheckpointCfg(delta_time=2), # Creates a checkpoint every two seconds (in simulation time)
)

and here is how to resume an experiment from a checkpoint

using Mavi.Experiment

experiment = load_experiment("path-to-experiment")
run_experiment(experiment)