adding functions to save joint expert states and actions for repeated use in metrics, adding option to flatten loaded trajectories, adding class to not shuffle trajectories when saving experts to make sure it lines up with the joint states. checked that it does

scratch/arec/intersimple/commands.txt

@@ -1 +1,22 @@
 python -m render_options --model_name='gail_options_image_mid_wcollision' --env='NRasterizedRoute' --options=True --width=36 --height=36 --m_per_px=2 --agent=50 --stop_on_collision=False
+
+import torch, os
+from src.data import load_experts
+folder = 'expert_data/DR_USA_Roundabout_FT/track0000'
+single_agent = os.path.join(folder, 'expert.pkl')
+multi_agent = os.path.join(folder,'joint_expert_states.pt')
+multi_agent_actions = os.path.join(folder,'joint_expert_actions.pt')
+demonstrations = load_experts([single_agent], flatten=False)
+demonstrations[0].__dict__.keys()
+len(demonstrations[0].obs)
+single_agent_lengths = [len(demonstration.obs) for demonstration in demonstrations]
+states = torch.load(multi_agent)
+actions = torch.load(multi_agent_actions)
+multi_agent_lengths = [sum(~torch.isnan(states[:,i,0])).item() for i in range(states.shape[1])]
+
+single_agent_actions = [demonstration.acts for demonstration in demonstrations]
+multi_agent_actions = [actions[~torch.isnan(actions[:,i,0])] for i in range(actions.shape[1])]
+
+import pickle
+with open(single_agent, "rb") as f:
+    new_trajectories = pickle.load(f)

src/data/__init__.py

@@ -1 +1 @@
-from src.data.expert import demonstrations, load_experts, process_experts 
+from src.data.expert import single_agent_demonstrations, multi_agent_demonstrations, NoShuffleRNG, load_experts, process_experts 

src/data/expert.py

@@ -1,3 +1,4 @@
+import intersim
 from intersim.envs.intersimple import Intersimple
 from stable_baselines3.common.policies import BasePolicy
 import gym
@@ -9,6 +10,7 @@ from stable_baselines3.common.vec_env.dummy_vec_env import DummyVecEnv
 from imitation.data.wrappers import RolloutInfoWrapper
 import copy
 import os
+import numpy as np
 
 class IntersimExpert(BasePolicy):
     
@@ -100,24 +102,40 @@ def save_video(env, expert):
         env.render()
     env.close()
 
-def load_experts(expert_files):
+class NoShuffleRNG(np.random.RandomState):
+    """
+    A np.random.RandomState rng that doesn't shuffle inputs (for imitation.rollout)
+    """
+    def __init__(self):
+        super().__init__()
+
+    def shuffle(self, x):
+        return x
+
+def load_experts(expert_files, flatten = True):
     """
     Load expert trajectories from files and combine their transitions into a single RB
 
     Args:
         expert_files (list): list of expert file strings
+        flatten (bool): whether to flatten trajectory info
     Returns:
         transitions (list): list of combined expert episode transitions 
     """
-    trajectories = []
+    transitions = []
     for file in tqdm(expert_files):
         with open(file, "rb") as f:
             new_trajectories = pickle.load(f)
-        trajectories += new_trajectories
-    transitions = rollout.flatten_trajectories(trajectories)
+        transitions += new_trajectories
+    if flatten:
+        transitions = rollout.flatten_trajectories(transitions)
     return transitions
 
-def demonstrations(expert='NormalizedIntersimpleExpert', env='NRasterizedRouteIncrementingAgent', path=None, min_timesteps=None, min_episodes=None, video=False, env_args={}, policy_args={}):
+def single_agent_demonstrations(expert='NormalizedIntersimpleExpert', 
+                                env='NRasterizedRouteIncrementingAgent', 
+                                path=None, min_timesteps=None, 
+                                min_episodes=None, video=False,
+                                env_args={}, policy_args={}):
     """Rollout and save expert demos.
     
     Usage:
@@ -154,13 +172,39 @@ def demonstrations(expert='NormalizedIntersimpleExpert', env='NRasterizedRouteIn
             min_timesteps=min_timesteps,
             min_episodes=min_episodes,
         )
+    
     rollout.rollout_and_save(
         path=path,
         policy=venv_policy,
         venv=venv,
-        sample_until=suntil
+        sample_until=suntil,
+        rng=NoShuffleRNG()
     )
 
+def multi_agent_demonstrations(expert='IntersimExpert',path=None, env_args={}, policy_args={}):
+    """
+    Run and save the `intersim' multiagent environment demonstration
+    
+    Args: 
+        expert (class): class of multi-agent expert
+        path (str): path to store output data
+        env_args (dict): dictionary of kwargs when instantiating environment class
+        policy_args (dict): dictionary of kwargs when instantiating Expert policy
+    """
+    if path is None:
+        raise('No path specified')
+
+    env = gym.make('intersim:intersim-v0',**env_args)
+    Expert = globals()[expert] 
+    policy = Expert(env, **policy_args)
+
+    s, done = env.reset(), False
+    env.render(mode='file')
+    while not done:
+        _,_,done,_ = env.step(policy.predict()[0])
+        env.render(mode='file')
+    env.close(filestr=path)
+
 def process_experts(filename:str='expert.pkl', 
                     locs:list=None, 
                     tracks:list=None,
@@ -190,22 +234,33 @@ def process_experts(filename:str='expert.pkl',
             iloc = intersim.LOCATIONS.index(loc)
 
             it_env_args = copy.deepcopy(env_args)
-            it_env_args.update({
+            env_loc_args = {
                 'loc':iloc,
                 'track':track,
-            })
+            }
+            it_env_args.update(env_loc_args)
             out_folder = os.path.join('expert_data',loc, 'track%04i'%(track))
             if not os.path.isdir(out_folder):
                 os.makedirs(out_folder) 
             it_path = os.path.join(out_folder,filename)
 
-            demonstrations(
+            # Multi-Agent demonstrations
+            it_ma_path = os.path.join(out_folder,'joint_expert')
+            multi_agent_demonstrations(
+                expert='IntersimExpert',
+                path=it_ma_path,
+                env_args=env_loc_args,
+                policy_args=expert_args
+            )
+            # Single-Agent POV Demonstrations
+            single_agent_demonstrations(
                 expert=expert_class, 
                 env=env_class, 
                 path=it_path, 
                 env_args=it_env_args, 
                 policy_args=expert_args,
                 )
+            
             pbar.update(1)
     pbar.close()