class Main:
train_results = "Train results 11042020F"
def __init__(self):
self.environment = Environment(NUMBER_OF_DRONES, NUMBER_OF_HUMANS)
self.agent = PolicyAgent() # create agent with default parameters
#self.agent.load_model("saved_agents\\policy_agent.h5")
# generating plots
def generate_plot(self, data: list):
fig, ax = plt.subplots()
ax.plot(data, 'r', label=f"{self.train_results}")
ax.set(xlabel="Episode", ylabel="Distance", title="")
ax.grid()
plt.legend(loc='upper left')
plt.draw()
fig.savefig(f'results\\{self.train_results}.png', dpi=1200)
# testing phase->
def training_loop(self):
count = 0
# iterate on epochs
for episode in range(TOTAL_EPOCHS):
print("Epoch # " + str(episode + 1))
# reinitialize the envoirment
state = self.environment.reset()
# set agent drone to update location
self.agent.drone = self.environment.drones[0]
# maximum number of steps to move
for step in range(MAX_STEPS):
# agent get the state and decide move, more detail in available in the function
action = self.agent.make_move(self.environment, state)
# make move function has updated the drone location now render the env
self.environment.render()
# check for the reward, get next location and if we reached to goal state
state_, reward, done = self.environment.get_info(
self.agent.drone)
self.agent.addState(state)
self.agent.store_experience(
[state, action, reward, state_, done])
state = state_
# if took specified number of steps it will update policy states
if len(self.agent.memory) > self.agent.min_memory_size:
self.agent.update(self.environment)
if done:
break
# update learning rate
self.agent.update_epsilon()
# after every iteration store the model
if (episode + 1) % (NUMBER_OF_EPOCHS) == 0:
print("saving model...")
self.agent.save_model(
"saved_agents\\policy_agent_{}.h5".format(count))
count += 1
# turn off the env
self.environment.close()
class Main:
train_results = "Train results 11042020F"
def __init__(self):
self.environment = Environment(NUMBER_OF_DRONES, NUMBER_OF_HUMANS)
self.agent = DDQNAgent(learning_rate=0.001,
epsilon=0.9,
epsilon_decay=0.99,
gamma=0.8,
batch_size=64,
buffer_size=10000,
min_memory_size=500,
tau=0.1)
self.agent.model = self.agent.create_model()
self.agent.target_model = self.agent.create_model()
# generating plots
def generate_plot(self, data: list):
fig, ax = plt.subplots()
ax.plot(data, 'r', label=f"{self.train_results}")
ax.set(xlabel="Episode", ylabel="Distance", title="")
ax.grid()
plt.legend(loc='upper left')
plt.draw()
fig.savefig(f'results\\{self.train_results}.png', dpi=1200)
# testing phase->
def training_loop(self):
for episode in range(NUMBER_OF_EPOCHS):
print(episode)
state = self.environment.reset()
self.agent.drone = self.environment.drones[0]
for step in range(MAX_STEPS):
action = self.agent.make_move(self.environment, state)
self.environment.render()
state_, reward, done = self.environment.get_info(self.agent.drone)
self.agent.store_experience([state, action, reward, state_, done])
state = state_
if len(self.agent.memory) > self.agent.min_memory_size:
self.agent.update(self.environment)
self.agent.update_target_weights()
if done:
break
self.agent.update_epsilon()
self.generate_plot(self.environment.drones[0].distance_history)
# self.agent.generate_loss()
self.agent.save_model()
self.environment.close()
# self.testing_loop()
def testing_loop(self):
model = keras.models.load_model("saved_agents\\agent.h5")