def create_agent_66_experiment_runner(num_dcs, num_customers, dcs_per_customer,
demand_mean, demand_var, num_commodities,
orders_per_day, num_steps):
physical_network = PhysicalNetwork(
num_dcs,
num_customers,
dcs_per_customer,
demand_mean,
demand_var,
num_commodities,
)
order_generator = ActualOrderGenerator(physical_network, orders_per_day)
generator = DirichletInventoryGenerator(physical_network)
environment_parameters = EnvironmentParameters(physical_network,
order_generator, generator,
num_steps)
env = ShippingFacilityEnvironment(environment_parameters)
agent = AgentHighest(env)
return ExperimentRunner(order_generator,
generator,
agent,
env,
experiment_name="randomvalid_validation")
def build_network_flow_env_parameters(environment_config, episode_length,
order_gen: str):
physical_network = PhysicalNetwork(
num_dcs=environment_config['num_dcs'],
num_customers=environment_config['num_customers'],
dcs_per_customer=environment_config['dcs_per_customer'],
demand_mean=environment_config['demand_mean'],
demand_var=environment_config['demand_var'],
big_m_factor=environment_config['big_m_factor'],
num_commodities=environment_config['num_commodities'],
)
if order_gen == 'original': # The original is independent means for each product customer.
order_generator = ActualOrderGenerator(
physical_network, environment_config['orders_per_day'])
elif order_gen == 'biased': # biased is more skewed and there's correlations in products.
order_generator = BiasedOrderGenerator( # todo make this parameterized with a factory or something.
physical_network, environment_config['orders_per_day'])
else:
raise NotImplementedError("alternatives are original and biased")
generator = DirichletInventoryGenerator(physical_network)
environment_parameters = EnvironmentParameters(physical_network,
order_generator, generator,
episode_length)
return environment_parameters
def run_with_params(num_dcs, num_customers, dcs_per_customer, demand_mean,
demand_var, num_commodities, orders_per_day, num_episodes):
physical_network = PhysicalNetwork(num_dcs, num_customers,
dcs_per_customer, demand_mean,
demand_var, num_commodities)
# order_generator = NaiveOrderGenerator(num_dcs, num_customers, orders_per_day)
order_generator = ActualOrderGenerator(physical_network, orders_per_day)
generator = NaiveInventoryGenerator()
environment_parameters = EnvironmentParameters(physical_network,
order_generator, generator,
num_episodes)
env = ShippingFacilityEnvironment(environment_parameters)
agent = RandomAgent(env.action_space)
obs = env.reset()
reward = 0
done = False
print("=========== starting episode loop ===========")
print("Initial environment: ")
env.render()
while not done:
action = agent.get_action(obs, reward, done)
#print(f"Agent is taking action: {action}")
# the agent observes the first state and chooses an action
# environment steps with the agent's action and returns new state and reward
obs, reward, done, info = env.step(action)
#print(f"Got reward {reward} done {done}")
# Render the current state of the environment
env.render()
if done:
print("===========Environment says we are DONE ===========")
def run_with_params(num_dcs, num_customers, dcs_per_customer, demand_mean,
demand_var, num_commodities, orders_per_day, num_steps):
physical_network = PhysicalNetwork(
num_dcs,
num_customers,
dcs_per_customer,
demand_mean,
demand_var,
num_commodities,
)
# order_generator = NaiveOrderGenerator(num_dcs, num_customers, orders_per_day)
order_generator = ActualOrderGenerator(physical_network, orders_per_day)
generator = DirichletInventoryGenerator(physical_network)
environment_parameters = EnvironmentParameters(physical_network,
order_generator, generator,
num_steps)
env = ShippingFacilityEnvironment(environment_parameters)
agent = RandomAgent(env)
obs = env.reset()
reward = 0
done = False
print("=========== starting episode loop ===========")
print("Initial environment: ")
env.render()
actions = []
episode_rewards = []
#demands_per_k = np.zeros((num_commodities,num_steps))
#inventory_at_t = np.zeros((num_commodities,num_steps)) #todo llenar estos eventualmente
while not done:
action = agent.act(obs, reward, done)
# print(f"Agent is taking action: {action}")
# the agent observes the first state and chooses an action
# environment steps with the agent's action and returns new state and reward
obs, reward, done, info = env.step(action)
# print(f"Got reward {reward} done {done}")
# Render the current state of the environment
env.render()
actions.append(action)
episode_rewards.append(reward)
if done:
print("===========Environment says we are DONE ===========")
return actions, episode_rewards
from shipping_allocation.envs.network_flow_env import EnvironmentParameters, NaiveOrderGenerator, NaiveInventoryGenerator
from gym.envs.registration import register
from network.PhysicalNetwork import PhysicalNetwork
register(id='shipping-v0',
entry_point='shipping_allocation.envs:ShippingFacilityEnvironment',
kwargs={
'environment_parameters':
EnvironmentParameters(PhysicalNetwork(5, 5, 2, 100, 20),
NaiveOrderGenerator(5, 5, 2),
NaiveInventoryGenerator(), 5)
})