def render(self, file_name):
fig, axs = plt.subplots(2, 1, figsize=(25, 10))
x = np.linspace(0, self.episod_len, self.episod_len)
_agent_list = []
_step_balances_idx = []
for i, f in enumerate(self.facility_names):
if (f.startswith('SKUStoreUnit')
or f.startswith('OuterSKUStoreUnit')
) and Utils.is_consumer_agent(f):
_agent_list.append(f)
_step_balances_idx.append(i)
_step_balances = [
self.step_balances[0, :, i] for i in _step_balances_idx
]
# axs[0].set_title('Global balance')
# axs[0].plot(x, self.global_balances.T)
axs[0].set_title('Cumulative Sum of Balance')
axs[0].plot(x, np.cumsum(np.sum(_step_balances, axis=0)))
axs[1].set_title('Reward Breakdown by Agent (One Episod)')
axs[1].plot(x, np.cumsum(_step_balances, axis=0).T)
axs[1].legend(_agent_list, loc='upper left')
fig.savefig(file_name)
def _actions_to_control(self, facility, actions):
control = FacilityCell.Control(unit_price=0,
production_rate=0,
consumer_product_id=0,
consumer_source_id=0,
consumer_quantity=0,
consumer_vlt=0)
consumer_action_list = Utils.get_consumer_action_space()
if isinstance(facility, FacilityCell):
return control
for agent_id, action in actions:
# action = np.array(action).flatten()
if Utils.is_producer_agent(agent_id):
if isinstance(facility, SKUSupplierUnit):
control.production_rate = facility.sku_info[
'production_rate']
if Utils.is_consumer_agent(agent_id):
product_id = facility.bom.output_product_id
control.consumer_product_id = product_id
if facility.consumer.sources is not None:
source = facility.consumer.sources[0]
control.consumer_vlt = source.sku_info['vlt']
control.consumer_source_id = 0 # action[0]
control.consumer_quantity = int(consumer_action_list[action] *
facility.get_sale_mean())
return control
def get_retailer_profit(self):
_agent_list = []
_step_balances_idx = []
for i, f in enumerate(self.facility_names):
if f.startswith('RetailerCell') and Utils.is_consumer_agent(f):
_agent_list.append(f)
_step_balances_idx.append(i)
_step_balances = [
self.step_balances[0, :, i] for i in _step_balances_idx
]
return np.sum(_step_balances)
def load_policy(agent_id):
if Utils.is_producer_agent(agent_id):
return ProducerBaselinePolicy(
env.observation_space, env.action_space_producer,
BaselinePolicy.get_config_from_env(env))
elif Utils.is_consumer_agent(agent_id):
return ConsumerMinMaxPolicy(
env.observation_space, env.action_space_consumer,
BaselinePolicy.get_config_from_env(env))
else:
raise Exception(f'Unknown agent type {agent_id}')