def test_agent(logfile):
display_on = True
# Create a random seed, which will define the environment
random_seed = int(time.time())
np.random.seed(random_seed)
# Create a random environment
environment = Environment(magnification=500)
# Create an agent
agent = Agent()
# Get the initial state
state = environment.init_state
# Determine the time at which training will stop, i.e. in 10 minutes (600 seconds) time
start_time = time.time()
end_time = start_time + 600
# Train the agent, until the time is up
while time.time() < end_time:
# If the action is to start a new episode, then reset the state
if agent.has_finished_episode():
state = environment.init_state
# Get the state and action from the agent
action = agent.get_next_action(state)
# Get the next state and the distance to the goal
next_state, distance_to_goal = environment.step(state, action)
# Return this to the agent
agent.set_next_state_and_distance(next_state, distance_to_goal)
# Set what the new state is
state = next_state
# Optionally, show the environment
if display_on:
environment.show(state)
# Test the agent for 100 steps, using its greedy policy
state = environment.init_state
has_reached_goal = False
for step_num in range(100):
action = agent.get_greedy_action(state)
next_state, distance_to_goal = environment.step(state, action)
# The agent must achieve a maximum distance of 0.03 for use to consider it "reaching the goal"
if distance_to_goal < 0.03:
has_reached_goal = True
break
state = next_state
# Print out the result
resultstr = (f"Reached goal in {step_num} steps." if has_reached_goal else
f"Did not reach goal. Final distance = {distance_to_goal}")
print(f"Result for maze {random_seed}: {resultstr}")
with open(logfile, "a") as f:
f.write(f"Result for maze {random_seed}: {resultstr}\n")
return has_reached_goal
while time.time() < end_time:
# If the action is to start a new episode, then reset the state
if agent.has_finished_episode():
state = environment.init_state
print("finished episode in time:", (time.time() - start_time) / 60)
# Get the state and action from the agent
action = agent.get_next_action(state)
# Get the next state and the distance to the goal
next_state, distance_to_goal = environment.step(state, action)
# Return this to the agent
agent.set_next_state_and_distance(next_state, distance_to_goal)
# Set what the new state is
state = next_state
# Optionally, show the environment
if display_on:
environment.show(state)
# Test the agent for 100 steps, using its greedy policy
state = environment.init_state
has_reached_goal = False
greedy_states = []
for step_num in range(100):
action = agent.get_greedy_action(state)
next_state, distance_to_goal = environment.step(state, action)
print("distance to goal:", distance_to_goal)
# The agent must achieve a maximum distance of 0.03 for use to consider it "reaching the goal"
if distance_to_goal < 0.03:
has_reached_goal = True
break
def train():
config_defaults = {
'batch_size': 400,
'learning_rate': 0.01,
'gamma': 0.95,
'epsilon_decay': 0.993,
'episode_length': 1000,
'network_update': 50
}
# Initialize a new wandb run
wandb.init(config=config_defaults)
# Config is a variable that holds and saves hyperparameters and inputs
config = wandb.config
# This determines whether the environment will be displayed on each each step.
# When we train your code for the 10 minute period, we will not display the environment.
display_on = False
# Create a random seed, which will define the environment
random_seed = int(time.time())
np.random.seed(random_seed)
# Create a random environment
environment = Environment(magnification=500)
# Create an agent
agent = Agent(config_defaults)
# Get the initial state
state = environment.init_state
# Determine the time at which training will stop, i.e. in 10 minutes (600 seconds) time
start_time = time.time()
end_time = start_time + 600
# Train the agent, until the time is up
while time.time() < end_time:
# If the action is to start a new episode, then reset the state
if agent.has_finished_episode():
state = environment.init_state
# Get the state and action from the agent
action = agent.get_next_action(state)
# Get the next state and the distance to the goal
next_state, distance_to_goal = environment.step(state, action)
wandb.log({"loss": distance_to_goal})
# Return this to the agent
agent.set_next_state_and_distance(next_state, distance_to_goal)
# Set what the new state is
state = next_state
# Optionally, show the environment
if display_on:
environment.show(state)
if agent.has_finished_episode():
wandb.log({"episode_end_distance": distance_to_goal})
# Test the agent for 100 steps, using its greedy policy
state = environment.init_state
has_reached_goal = False
for step_num in range(100):
action = agent.get_greedy_action(state)
next_state, distance_to_goal = environment.step(state, action)
# The agent must achieve a maximum distance of 0.03 for use to consider it "reaching the goal"
if distance_to_goal < 0.03:
has_reached_goal = True
wandb.log({"goal_reached": has_reached_goal})
break
state = next_state
# Print out the result
if has_reached_goal:
print('Reached goal in ' + str(step_num) + ' steps.')
else:
print('Did not reach goal. Final distance = ' + str(distance_to_goal))
wandb.log({"distance_to_goal": distance_to_goal})
