}
}))
tree = makeTree(setToConstantMatrix(my_dec, COOPERATED))
world.setDynamics(my_dec, action, tree)
# defines payoff matrices (equal to both agents)
agent1.setReward(get_reward_tree(agent1, agents_dec[0], agents_dec[1]), 1)
agent2.setReward(get_reward_tree(agent2, agents_dec[1], agents_dec[0]), 1)
# define order
my_turn_order = [{agent1.name, agent2.name}]
world.setOrder(my_turn_order)
# add true mental model of the other to each agent
world.setMentalModel(agent1.name, agent2.name,
Distribution({agent2.get_true_model(): 1}))
world.setMentalModel(agent2.name, agent1.name,
Distribution({agent1.get_true_model(): 1}))
for h in range(MAX_HORIZON + 1):
logging.info('====================================')
logging.info(f'Horizon {h}')
# set horizon (also to the true model!) and reset decisions
for i in range(len(agents)):
agents[i].setHorizon(h)
agents[i].setHorizon(h, agents[i].get_true_model())
world.setFeature(agents_dec[i], NOT_DECIDED, recurse=True)
for t in range(NUM_STEPS):
# add mental model of the other for each agent
world.setMentalModel(agent1.name, agent2.name,
Distribution({get_fake_model_name(agent2): 1}))
world.setMentalModel(agent2.name, agent1.name,
Distribution({get_fake_model_name(agent1): 1}))
# 'hides' right actions from models by setting them illegal
# (therefore agents should always choose right because they think the other will choose left)
set_illegal_action(agent1, rights[0], [get_fake_model_name(agent1)])
set_illegal_action(agent2, rights[1], [get_fake_model_name(agent2)])
# # ** unnecessary / just for illustration **: set left actions legal for both the agents and their models
# set_legal_action(agent1, lefts[0], [agent1.get_true_model(), get_fake_model_name(agent1)])
# set_legal_action(agent2, lefts[1], [agent2.get_true_model(), get_fake_model_name(agent2)])
agent1.resetBelief(model=agent1.get_true_model())
agent1.resetBelief(model=get_fake_model_name(agent1))
agent2.resetBelief(model=agent2.get_true_model())
agent2.resetBelief(model=get_fake_model_name(agent2))
for t in range(NUM_STEPS):
# reset decision
for a in range(len(agents)):
world.setFeature(sides[a], NOT_DECIDED, recurse=True)
logging.info('====================================')
logging.info(f'Step {t}')
step = world.step()
for a in range(len(agents)):
logging.info(
f'{agents[a].name}: {world.getFeature(sides[a], unique=True)}')