def learn_evolutionary():
base_agent = MiniSoccerAgent(rl.FeatureSet([]))
sample_state = base_agent.environment.generate_start_state()
state_vars = sample_state.state_variables
featurizer_retile = rl.FeaturizerRetile(state_vars)
featurizer_interaction = rl.FeaturizerInteraction(state_vars)
featurizer_angle = rl.FeaturizerAngle(state_vars)
featurizer_dist = rl.FeaturizerDist(state_vars)
featurizer_dist_x = rl.FeaturizerDistX(state_vars)
featurizer_dist_y = rl.FeaturizerDistY(state_vars)
featurizer_flag = rl.FeaturizerFlag(state_vars)
featurizer_point_xy = rl.FeaturizerPointXY(state_vars)
featurizer_point_x = rl.FeaturizerPointX(state_vars)
featurizer_point_y = rl.FeaturizerPointY(state_vars)
featurizers_map = [
(0.12, featurizer_retile),
(0.15, featurizer_interaction),
(0.10, featurizer_flag),
(0.16, featurizer_angle),
(0.12, featurizer_dist),
(0.09, featurizer_dist_x),
(0.09, featurizer_dist_y),
(0.07, featurizer_point_xy),
(0.05, featurizer_point_x),
(0.05, featurizer_point_y)
]
# featurizers_map = [(0.15, featurizer_retile),
# (0.10, featurizer_interaction),
# (0.10, featurizer_flag),
# (0.20, featurizer_angle),
# (0.15, featurizer_dist),
# (0.10, featurizer_dist_x),
# (0.10, featurizer_dist_y),
# (0.10, featurizer_point_xy),
# (0.0, featurizer_point_x),
# (0.0, featurizer_point_y)
# ]
arbitrator = rl.ArbitratorEvolutionary(base_agent, featurizers_map,
NUM_GENERATIONS, POPULATION_SIZE,
NUM_GENERATION_EPISODES, NUM_CHAMPION_TRIALS,
NUM_BEST_CHAMPION_EPISODES, NUM_BEST_CHAMPION_TRIALS,
rl.DEFAULT_ETA)
arbitrator.run(MAX_STEPS)
def learn_evolutionary():
base_agent = MiniSoccerAgent(rl.FeatureSet([]))
sample_state = base_agent.environment.generate_start_state()
state_vars = sample_state.state_variables
featurizer_retile = rl.FeaturizerRetile(state_vars)
featurizer_interaction = rl.FeaturizerInteraction(state_vars)
featurizer_angle = rl.FeaturizerAngle(state_vars)
featurizer_dist = rl.FeaturizerDist(state_vars)
featurizer_dist_x = rl.FeaturizerDistX(state_vars)
featurizer_dist_y = rl.FeaturizerDistY(state_vars)
featurizer_flag = rl.FeaturizerFlag(state_vars)
featurizer_point_xy = rl.FeaturizerPointXY(state_vars)
featurizer_point_x = rl.FeaturizerPointX(state_vars)
featurizer_point_y = rl.FeaturizerPointY(state_vars)
featurizers_map = [(0.12, featurizer_retile),
(0.15, featurizer_interaction), (0.05, featurizer_flag),
(0.20, featurizer_angle), (0.08, featurizer_dist),
(0.08, featurizer_dist_x), (0.08, featurizer_dist_y),
(0.08, featurizer_point_xy), (0.08, featurizer_point_x),
(0.08, featurizer_point_y)]
featurizers_map = [(0.15, featurizer_retile),
(0.10, featurizer_interaction), (0.10, featurizer_flag),
(0.20, featurizer_angle), (0.15, featurizer_dist),
(0.10, featurizer_dist_x), (0.10, featurizer_dist_y),
(0.10, featurizer_point_xy), (0.0, featurizer_point_x),
(0.0, featurizer_point_y)]
sum_prob = 0.0
for (prob, featurizer) in featurizers_map:
sum_prob += prob
print "Initialized %d featurizers, sum of selection probabilities: %f" % (
len(featurizers_map), sum_prob)
arbitrator = rl.ArbitratorEvolutionary(base_agent, featurizers_map,
NUM_GENERATIONS, POPULATION_SIZE,
GENERATION_EPISODES)
arbitrator.execute(MAX_STEPS)
def learn_evolutionary():
base_agent = MiniSoccerAgent(rl.FeatureSet([]))
sample_state = base_agent.environment.generate_start_state()
state_vars = sample_state.state_variables
featurizer_retile = rl.FeaturizerRetile(state_vars)
featurizer_angle = rl.FeaturizerAngle(state_vars)
featurizer_dist = rl.FeaturizerDist(state_vars)
featurizer_dist_x = rl.FeaturizerDistX(state_vars)
featurizer_dist_y = rl.FeaturizerDistY(state_vars)
featurizer_flag = rl.FeaturizerFlag(state_vars)
featurizer_point2d = rl.FeaturizerPoint2D(state_vars)
featurizers_map = [(0.20, featurizer_retile), (0.35, featurizer_dist),
(0.50, featurizer_dist_x), (0.65, featurizer_dist_y),
(0.80, featurizer_angle), (0.90, featurizer_point2d),
(1.00, featurizer_flag)]
arbitrator = rl.ArbitratorEvolutionary(base_agent, featurizers_map,
NUM_GENERATIONS, POPULATION_SIZE,
GENERATION_EPISODES)
arbitrator.execute(MAX_STEPS)
def learn_evolutionary():
base_agent = KeepAwayAgent(rl.FeatureSet([]))
sample_state = base_agent.environment.generate_start_state()
state_vars = sample_state.state_variables
# retile_featurizer = rl.FeaturizerRetile(state_vars)
# angle_featurizer = rl.FeaturizerAngle(state_vars)
# dist_featurizer = rl.FeaturizerDist(state_vars)
#
# featurizers_map = [(0.2, retile_featurizer),
# (0.4, angle_featurizer),
# (0.4, dist_featurizer)]
featurizer_retile = rl.FeaturizerRetile(state_vars)
featurizer_interaction = rl.FeaturizerInteraction(state_vars)
featurizer_angle = rl.FeaturizerAngle(state_vars)
featurizer_dist = rl.FeaturizerDist(state_vars)
featurizer_dist_x = rl.FeaturizerDistX(state_vars)
featurizer_dist_y = rl.FeaturizerDistY(state_vars)
# featurizer_flag = rl.FeaturizerFlag(state_vars)
featurizer_point_xy = rl.FeaturizerPointXY(state_vars)
featurizer_point_x = rl.FeaturizerPointX(state_vars)
featurizer_point_y = rl.FeaturizerPointY(state_vars)
featurizers_map = [(0.16, featurizer_retile),
(0.15, featurizer_interaction),
(0.16, featurizer_angle), (0.13, featurizer_dist),
(0.11, featurizer_dist_x), (0.11, featurizer_dist_y),
(0.08, featurizer_point_xy), (0.05, featurizer_point_x),
(0.05, featurizer_point_y)]
arbitrator = rl.ArbitratorEvolutionary(
base_agent, featurizers_map, NUM_GENERATIONS, POPULATION_SIZE,
NUM_GENERATION_EPISODES, NUM_CHAMPION_TRIALS,
NUM_BEST_CHAMPION_EPISODES, NUM_BEST_CHAMPION_TRIALS, rl.DEFAULT_ETA)
arbitrator.run()