diff --git a/scoring/generate_tests.py b/scoring/generate_tests.py
index 6839b039c42f08661196f05a3e1e8e899b5d8f26..edec302c305670134c1d594b1c89fd4c222e3a08 100644
--- a/scoring/generate_tests.py
+++ b/scoring/generate_tests.py
@@ -2,7 +2,7 @@ import time
import numpy as np
-from utils.misc_utils import create_testsfiles
+from utils.misc_utils import create_testfiles
with open('parameters.txt', 'r') as inf:
parameters = eval(inf.read())
@@ -21,5 +21,4 @@ test_idx = 0
for test_nr in parameters:
current_parameters = parameters[test_nr]
- create_testsfiles(current_parameters,test_nr)
-
+ create_testfiles(current_parameters, test_nr, nr_trials_per_test=100)
diff --git a/scoring/score_test.py b/scoring/score_test.py
index 1e46f080d8208af896134b1e6cacfdd476bf6b3c..79f0ee6a21590e021daa732afb7e2d5e8f68ee05 100644
--- a/scoring/score_test.py
+++ b/scoring/score_test.py
@@ -2,10 +2,13 @@ import time
import numpy as np
import torch
-from utils.misc_utils import run_test,run_test_sequential
-from dueling_double_dqn import Agent
-from sequential_agent.simple_order_agent import OrderedAgent
-with open('parameters.txt','r') as inf:
+from flatland.envs.observations import TreeObsForRailEnv
+from flatland.envs.predictions import ShortestPathPredictorForRailEnv
+
+from torch_training.dueling_double_dqn import Agent
+from utils.misc_utils import run_test
+
+with open('parameters.txt', 'r') as inf:
parameters = eval(inf.read())
# Parameter initialization
@@ -23,29 +26,26 @@ test_times = []
test_dones = []
sequential_agent_test = False
+# Load your agent
+agent = Agent(state_size, action_size, "FC", 0)
+agent.qnetwork_local.load_state_dict(torch.load('../torch_training/Nets/avoid_checkpoint60000.pth'))
-# Load agent
-if sequential_agent_test:
- agent = OrderedAgent()
-else:
- agent = Agent(state_size, action_size, "FC", 0)
- agent.qnetwork_local.load_state_dict(torch.load('./Nets/avoid_checkpoint60000.pth'))
+# Load the necessary Observation Builder and Predictor
+predictor = ShortestPathPredictorForRailEnv()
+observation_builder = TreeObsForRailEnv(max_depth=tree_depth, predictor=predictor)
start_time_scoring = time.time()
score_board = []
for test_nr in parameters:
current_parameters = parameters[test_nr]
- if sequential_agent_test:
- test_score, test_dones, test_time = run_test_sequential(current_parameters, agent, test_nr=test_nr, tree_depth=1)
-
- else:
- test_score, test_dones, test_time = run_test(current_parameters, agent, test_nr=test_nr,tree_depth=3)
+ test_score, test_dones, test_time = run_test(current_parameters, agent, observation_builder=observation_builder,
+ test_nr=test_nr, nr_trials_per_test=10)
print('{} score was {:.3f} with {:.2f}% environments solved. Test took {:.2f} Seconds to complete.\n'.format(
test_nr,
np.mean(test_score), np.mean(test_dones) * 100, test_time))
- score_board.append([np.mean(test_score), np.mean(test_dones) * 100, test_time])
+ score_board.append([np.mean(test_score), np.mean(test_dones) * 100, test_time])
print('---------')
print(' RESULTS')
print('---------')
diff --git a/scoring/utils/misc_utils.py b/scoring/utils/misc_utils.py
index e4aab524164275b47601783bde9d6b0703098ca0..de30bad398c5c09493b8850f02f13326648b2278 100644
--- a/scoring/utils/misc_utils.py
+++ b/scoring/utils/misc_utils.py
@@ -35,46 +35,17 @@ def printProgressBar(iteration, total, prefix='', suffix='', decimals=1, length=
print('')
-class RandomAgent:
-
- def __init__(self, state_size, action_size):
- self.state_size = state_size
- self.action_size = action_size
-
- def act(self, state, eps=0):
- """
- :param state: input is the observation of the agent
- :return: returns an action
- """
- return np.random.choice(np.arange(self.action_size))
-
- def step(self, memories):
- """
- Step function to improve agent by adjusting policy given the observations
-
- :param memories: SARS Tuple to be
- :return:
- """
- return
-
- def save(self, filename):
- # Store the current policy
- return
-
- def load(self, filename):
- # Load a policy
- return
-
-
-def run_test(parameters, agent, test_nr=0, tree_depth=3):
+def run_test(parameters, agent, observation_builder=None, observation_wrapper=None, test_nr=0, nr_trials_per_test=100):
# Parameter initialization
features_per_node = 9
start_time_scoring = time.time()
action_dict = dict()
- nr_trials_per_test = 100
+
print('Running {} with (x_dim,y_dim) = ({},{}) and {} Agents.'.format(test_nr, parameters[0], parameters[1],
parameters[2]))
-
+ if observation_builder == None:
+ print("No observation defined!")
+ return
# Reset all measurements
test_scores = []
test_dones = []
@@ -87,47 +58,39 @@ def run_test(parameters, agent, test_nr=0, tree_depth=3):
printProgressBar(0, nr_trials_per_test, prefix='Progress:', suffix='Complete', length=20)
for trial in range(nr_trials_per_test):
# Reset the env
- file_name = "./Envs/{}/Level_{}.pkl".format(test_nr, trial)
+ file_name = "./Tests/{}/Level_{}.pkl".format(test_nr, trial)
env = RailEnv(width=3,
height=3,
rail_generator=rail_from_file(file_name),
- obs_builder_object=TreeObsForRailEnv(max_depth=tree_depth, predictor=ShortestPathPredictorForRailEnv()),
+ obs_builder_object=observation_builder(),
number_of_agents=1,
)
obs = env.reset()
- agent_obs = [None] * env.get_num_agents()
- for a in range(env.get_num_agents()):
- data, distance, agent_data = split_tree(tree=np.array(obs[a]), num_features_per_node=9,
- current_depth=0)
- data = norm_obs_clip(data, fixed_radius=10)
- distance = norm_obs_clip(distance)
- agent_data = np.clip(agent_data, -1, 1)
- agent_obs[a] = np.concatenate((np.concatenate((data, distance)), agent_data))
+
+ if observation_wrapper is not None:
+ for a in range(env.get_num_agents()):
+ obs[a] = observation_wrapper(obs[a])
# Run episode
trial_score = 0
- max_steps = int(max_time_factor* (env.height + env.width))
+ max_steps = int(max_time_factor * (env.height + env.width))
for step in range(max_steps):
for a in range(env.get_num_agents()):
- action = agent.act(agent_obs[a], eps=0)
+ action = agent.act(obs[a], eps=0)
action_dict.update({a: action})
# Environment step
- next_obs, all_rewards, done, _ = env.step(action_dict)
+ obs, all_rewards, done, _ = env.step(action_dict)
for a in range(env.get_num_agents()):
- data, distance, agent_data = split_tree(tree=np.array(next_obs[a]),
- num_features_per_node=features_per_node,
- current_depth=0)
- data = norm_obs_clip(data, fixed_radius=10)
- distance = norm_obs_clip(distance)
- agent_data = np.clip(agent_data, -1, 1)
- agent_obs[a] = np.concatenate((np.concatenate((data, distance)), agent_data))
+ if observation_wrapper is not None:
+ obs[a] = observation_wrapper(obs[a])
trial_score += np.mean(all_rewards[a])
+
if done['__all__']:
break
test_scores.append(trial_score / max_steps)
@@ -138,9 +101,8 @@ def run_test(parameters, agent, test_nr=0, tree_depth=3):
return test_scores, test_dones, tot_test_time
-def create_testsfiles(parameters, test_nr=0):
+def create_testfiles(parameters, test_nr=0, nr_trials_per_test=100):
# Parameter initialization
- nr_trials_per_test = 100
print('Creating {} with (x_dim,y_dim) = ({},{}) and {} Agents.'.format(test_nr, parameters[0], parameters[1],
parameters[2]))
# Reset environment
@@ -159,7 +121,7 @@ def create_testsfiles(parameters, test_nr=0):
for trial in range(nr_trials_per_test):
# Reset the env
env.reset(True, True)
- env.save("./Envs/{}/Level_{}.pkl".format(test_nr, trial))
+ env.save("./Tests/{}/Level_{}.pkl".format(test_nr, trial))
printProgressBar(trial + 1, nr_trials_per_test, prefix='Progress:', suffix='Complete', length=20)
return
@@ -171,7 +133,7 @@ def render_test(parameters, test_nr=0, nr_examples=5):
print('Showing {} Level {} with (x_dim,y_dim) = ({},{}) and {} Agents.'.format(test_nr, trial, parameters[0],
parameters[1],
parameters[2]))
- file_name = "./Envs/{}/Level_{}.pkl".format(test_nr, trial)
+ file_name = "./Tests/{}/Level_{}.pkl".format(test_nr, trial)
env = RailEnv(width=1,
height=1,
@@ -210,7 +172,7 @@ def run_test_sequential(parameters, agent, test_nr=0, tree_depth=3):
printProgressBar(0, nr_trials_per_test, prefix='Progress:', suffix='Complete', length=20)
for trial in range(nr_trials_per_test):
# Reset the env
- file_name = "./Envs/{}/Level_{}.pkl".format(test_nr, trial)
+ file_name = "./Tests/{}/Level_{}.pkl".format(test_nr, trial)
env = RailEnv(width=3,
height=3,