Rollout scripts updates

rollout.py

@@ -15,9 +15,9 @@ import ray
 from ray.rllib.agents.registry import get_agent_class
 from ray.rllib.env import MultiAgentEnv
 from ray.rllib.env.base_env import _DUMMY_AGENT_ID
-from ray.rllib.evaluation.episode import _flatten_action
+# from ray.rllib.evaluation.episode import _flatten_action # ray 0.8.4
 from ray.rllib.policy.sample_batch import DEFAULT_POLICY_ID
-#from ray.rllib.utils.space_utils import flatten_to_single_ndarray
+from ray.rllib.utils.space_utils import flatten_to_single_ndarray # ray 0.8.5
 from ray.tune.utils import merge_dicts
 
 from utils.loader import load_envs, load_models
@@ -324,8 +324,8 @@ def rollout(agent,
         state_init = {p: m.get_initial_state() for p, m in policy_map.items()}
         use_lstm = {p: len(s) > 0 for p, s in state_init.items()}
         action_init = {
-            #p: flatten_to_single_ndarray(m.action_space.sample()) # ray 0.8.5
-            p: _flatten_action(m.action_space.sample()) # ray 0.8.4
+            p: flatten_to_single_ndarray(m.action_space.sample()) # ray 0.8.5
+            # p: _flatten_action(m.action_space.sample()) # ray 0.8.4
             for p, m in policy_map.items()
         }
     else:
@@ -389,8 +389,8 @@ def rollout(agent,
                             prev_action=prev_actions[agent_id],
                             prev_reward=prev_rewards[agent_id],
                             policy_id=policy_id)
-                    #a_action = flatten_to_single_ndarray(a_action)  # ray 0.8.5
-                    a_action = _flatten_action(a_action)  # tuple actions # ray 0.8.4
+                    a_action = flatten_to_single_ndarray(a_action)  # ray 0.8.5
+                    # a_action = _flatten_action(a_action)  # tuple actions # ray 0.8.4
                     action_dict[agent_id] = a_action
                     prev_actions[agent_id] = a_action
             action = action_dict
@@ -425,7 +425,7 @@ def rollout(agent,
 
         episode_score = sum(agents_score.values())
         simulation_rewards.append(episode_score)
-        simulation_rewards_normalized.append(episode_score / (episode_max_steps + episode_num_agents))
+        simulation_rewards_normalized.append(episode_score / (episode_max_steps * episode_num_agents))
         simulation_percentage_complete.append(float(len(agents_done)) / episode_num_agents)
         simulation_steps.append(episode_steps)
 

rollout.sh

+echo "===================="
+echo "APEX TREE OBS"
+echo "===================="
+
+# python rollout.py baselines/checkpoints/apex-tree-obs-small-v0-0/checkpoint_400/checkpoint-400 --run APEX --no-render --episodes 50 --env 'flatland_sparse' --config '{"env_config": {"seed":1000000000,"generator": "sparse_rail_generator", "generator_config": "small_v0", "observation": "tree", "observation_config": {"max_depth": 2, "shortest_path_max_depth": 30}}, "model": {"fcnet_activation": "relu", "fcnet_hiddens": [256, 256], "vf_share_layers": "True"}}'
+
+# python rollout.py baselines/checkpoints/apex_tree_obs_small_v0-1/checkpoint_400/checkpoint-400 --run APEX --no-render --episodes 50 --env 'flatland_sparse' --config '{"env_config": {"seed":1000000000,"generator": "sparse_rail_generator", "generator_config": "small_v0", "observation": "tree", "observation_config": {"max_depth": 2, "shortest_path_max_depth": 30}}, "model": {"fcnet_activation": "relu", "fcnet_hiddens": [256, 256], "vf_share_layers": "True"}}'
+
+# python rollout.py baselines/checkpoints/apex_tree_obs_small_v0-2/checkpoint_250/checkpoint-250 --run APEX --no-render --episodes 50 --env 'flatland_sparse' --config '{"env_config": {"seed":1000000000,"generator": "sparse_rail_generator", "generator_config": "small_v0", "observation": "tree", "observation_config": {"max_depth": 2, "shortest_path_max_depth": 30}}, "model": {"fcnet_activation": "relu", "fcnet_hiddens": [256, 256], "vf_share_layers": "True"}}'
\ No newline at end of file