def main(args):
env = ContinuousEnv3()
g = model_utils.BalancedMLP([1], [8, 16], 0.001, 32, 0.0, verbose=True)
def visualize_b(state_action_partition):
vis_utils.plot_background(env, show=False)
vis_utils.plot_decision_boundary(g,
env.STATE_ACTION_MAP.shape[1],
env.STATE_ACTION_MAP.shape[0],
show=False)
vis_utils.plot_state_action_partition(state_action_partition,
show=True)
def visualize_sb(state_partition):
vis_utils.plot_background(env, show=False)
vis_utils.plot_state_partition(state_partition, show=True)
experience = gather_experience(env, args.num_experience)
h**o = online_homomorphism_g.OnlineHomomorphismG(
experience,
g,
sample_actions,
args.state_action_slit_threshold,
args.state_split_threshold,
20,
visualize_b=visualize_b,
visualize_sb=visualize_sb)
h**o.partition_iteration()
hits, total = evaluation.overlap(env, list(h**o.partition))
print("{:.2f}% accuracy ({:d}/{:d})".format((hits / total) * 100, hits,
total))
def test_overlap_single_block(self):
block1 = [[0.1, 0.1], [0.4, 1.2], [1.2, 0.1], [1.9, 1.9]]
partition = [block1]
hits = evaluation.overlap(self.MockEnv, partition)
self.assertEqual(hits, 2)
def test_overlap_two_blocks(self):
block1 = [[0.1, 0.1], [0.4, 0.2]]
block2 = [[1.5, 0.5], [1.3, 1.7]]
partition = [block1, block2]
hits = evaluation.overlap(self.MockEnv, partition)
self.assertEqual(hits, 3)
def test_overlap_total_match(self):
block1 = [[0.1, 0.1], [0.4, 0.2]]
block2 = [[1.5, 0.5], [1.3, 0.7]]
block3 = [[1.7, 1.1], [1.9, 1.2]]
partition = [block1, block2, block3]
hits = evaluation.overlap(self.MockEnv, partition)
self.assertEqual(hits, 6)
def run(resolution):
env = ContinuousEnv3()
g = model_utils.BalancedMLP([1], [8, 16], 0.001, 32, 0.0, verbose=True)
experience = gather_experience(env, NUM_EXPERIENCE)
h**o = OnlineHomomorphismGDict(experience, g, sample_actions,
SPLIT_THRESHOLD, resolution, 20)
h**o.partition_iteration()
hits, total = evaluation.overlap(env, list(h**o.partition))
accuracy = hits / total
return accuracy
def run(num_experience, split_threshold, min_confidence):
env = ContinuousEnv3()
g = model_utils.BalancedMLP([1], [8, 16], 0.001, 32, 0.0, verbose=True)
experience = gather_experience(env, num_experience)
h**o = OnlineHomomorphismG(experience, g, sample_actions, split_threshold,
0, 20, min_confidence)
h**o.partition_iteration()
hits, total = evaluation.overlap(env, list(h**o.partition))
accuracy = hits / total
return accuracy
def run(num_experience, threshold_multiplier):
env = ContinuousEnv3()
g = model_utils.BalancedMLP([1], [8, 16], 0.001, 32, 0.0, verbose=True)
experience = gather_experience(env, num_experience)
h**o = OnlineHomomorphismGDict(experience, g, sample_actions, threshold_multiplier, MINIMUM_THRESHOLD,
OnlineHomomorphismGDict.RESOLVE_IGNORE, 20)
h**o.partition_iteration()
hits, total = evaluation.overlap(env, list(h**o.partition))
accuracy = hits / total
return accuracy
