Simulation

Simulation

Single Simulation

For simulating an adaptive test, first the item pool has to be initialized.

import adaptivetesting as adt
item_pool = adt.ItemPool.load_from_list(
    a = [0.42, 0.3, 1.5],
    b = [0.5, 0.9, 1.1]
)

Then, the test object can be created using the TestAssembler class. For reproducibility, a seed can be set.

test = adt.TestAssembler(
    item_pool = item_pool,
    simulation_id="test",
    participant_id="1",
    ability_estimator=adt.MLEstimator,
    true_ability_level=0,
    seed=123
)

The simulation itself is set up with the Simulation class which specifies the output format for the test results and the stopping criteria.

simulation = adt.Simulation(test, adt.ResultOutputFormat.CSV)

simulation.simulate(criterion=adt.StoppingCriterion.SE, value=0.4)
simulation.save_test_results()

Simulation Pool

For large scale simulations, the SimulationPool class can be used. For every single simulation, a test object has to be created which needs to be saved in a list.

tests = [
	adt.TestAssembler(
		item_pool=item_pool,
		simulation_id="example",
		participant_id="1",
		ability_estimator=adt.MLEstimator,
		item_selector=adt.maximum_information_criterion,
		true_ability_level=0,
		seed=123
	),
    adt.TestAssembler(
		item_pool=item_pool,
		simulation_id="example",
		participant_id="2",
		ability_estimator=adt.MLEstimator,
		item_selector=adt.maximum_information_criterion,
		true_ability_level=1,
		seed=123
	),
]

The simulation pool allows the tests to run sequentially without additional setup but also in parallel.

sim_pool = adt.SimulationPool(
	adaptive_tests=tests,
	test_result_output=adt.ResultOutputFormat.CSV,
	criterion=adt.StoppingCriterion.SE,
	value=0.4
)
sim_pool.start()

Depending on the operating system, the simulation pool uses either multithreading (on Windows) or multiprocessing (on other platforms) to run the simulation for each adaptive test. Note that parallel processing is not supported for the use in jupyter notebooks. For that, parallel has to be set to False.