Merge remote-tracking branch 'origin/main'

algorithms/utils.py

@@ -1,9 +1,16 @@
 import re
 import torch
+import numpy as np
 import yaml
 from pathlib import Path
 from salina import instantiate_class
 from salina import TAgent
+from salina.agents.gyma import (
+    AutoResetGymAgent,
+    _torch_type,
+    _format_frame,
+    _torch_cat_dict
+)
 
 
 def load_yaml_file(path: Path):
@@ -18,13 +25,86 @@ def add_env_props(cfg):
                              n_actions=env.action_space.n))
 
 
+
+
+AGENT_PREFIX = 'agent#'
+REWARD       =  'reward'
+CUMU_REWARD  = 'cumulated_reward'
+OBS          = 'env_obs'
+SEP          = '_'
+ACTION       = 'action'
+
+
+def access_str(agent_i, name, prefix=''):
+    return f'{prefix}{AGENT_PREFIX}{agent_i}{SEP}{name}'
+
+
+class AutoResetGymMultiAgent(AutoResetGymAgent):
+    def __init__(self, *args, **kwargs):
+        super(AutoResetGymMultiAgent, self).__init__(*args, **kwargs)
+
+    def per_agent_values(self, name, values):
+        return {access_str(agent_i, name): value
+                for agent_i, value in zip(range(self.n_agents), values)}
+
+    def _initialize_envs(self, n):
+        super()._initialize_envs(n)
+        n_agents_list = [self.envs[i].unwrapped.n_agents for i in range(n)]
+        assert all(n_agents == n_agents_list[0] for n_agents in n_agents_list), \
+            'All envs must have the same number of agents.'
+        self.n_agents = n_agents_list[0]
+
+    def _reset(self, k, save_render):
+        ret = super()._reset(k, save_render)
+        obs = ret['env_obs'].squeeze()
+        self.cumulated_reward[k] = [0.0]*self.n_agents
+        obs      = self.per_agent_values(OBS,  [_format_frame(obs[i]) for i in range(self.n_agents)])
+        cumu_rew = self.per_agent_values(CUMU_REWARD, torch.zeros(self.n_agents, 1).float().unbind())
+        rewards  = self.per_agent_values(REWARD,      torch.zeros(self.n_agents, 1).float().unbind())
+        ret.update(cumu_rew)
+        ret.update(rewards)
+        ret.update(obs)
+        for remove in ['env_obs', 'cumulated_reward', 'reward']:
+            del ret[remove]
+        return ret
+
+    def _step(self, k, action, save_render):
+        self.timestep[k] += 1
+        env = self.envs[k]
+        if len(action.size()) == 0:
+            action = action.item()
+            assert isinstance(action, int)
+        else:
+            action = np.array(action.tolist())
+        o, r, d, _ = env.step(action)
+        self.cumulated_reward[k] = [x+y for x, y in zip(r, self.cumulated_reward[k])]
+        observation = self.per_agent_values(OBS, [_format_frame(o[i]) for i in range(self.n_agents)])
+        if d:
+            self.is_running[k] = False
+        if save_render:
+            image = env.render(mode="image").unsqueeze(0)
+            observation["rendering"] = image
+        rewards           = self.per_agent_values(REWARD, torch.tensor(r).float().view(-1, 1).unbind())
+        cumulated_rewards = self.per_agent_values(CUMU_REWARD, torch.tensor(self.cumulated_reward[k]).float().view(-1, 1).unbind())
+        ret = {
+            **observation,
+            **rewards,
+            **cumulated_rewards,
+            "done": torch.tensor([d]),
+            "initial_state": torch.tensor([False]),
+            "timestep": torch.tensor([self.timestep[k]])
+        }
+        return _torch_type(ret)
+
+
 class CombineActionsAgent(TAgent):
-    def __init__(self, pattern=r'^agent\d_action$'):
+    def __init__(self):
         super().__init__()
-        self.pattern = pattern
+        self.pattern = fr'^{AGENT_PREFIX}\d{SEP}{ACTION}$'
 
     def forward(self, t, **kwargs):
         keys = list(self.workspace.keys())
         action_keys = sorted([k for k in keys if bool(re.match(self.pattern, k))])
         actions = torch.cat([self.get((k, t)) for k in action_keys], 0)
-        self.set((f'action', t), actions.unsqueeze(0))
+        actions = actions if len(action_keys) <= 1 else actions.unsqueeze(0)
+        self.set((f'action', t), actions)

environments/factory/__init__.py

@@ -1,4 +1,4 @@
-def make(env_name, n_agents=1, pomdp_r=2, max_steps=400, stack_n_frames=3):
+def make(env_name, pomdp_r=2, max_steps=400, stack_n_frames=3, n_agents=1,  individual_rewards=False):
     import yaml
     from pathlib import Path
     from environments.factory.combined_factories import DirtItemFactory
@@ -12,7 +12,8 @@ def make(env_name, n_agents=1, pomdp_r=2, max_steps=400, stack_n_frames=3):
     obs_props = ObservationProperties(render_agents=AgentRenderOptions.COMBINED,
                                       frames_to_stack=stack_n_frames, pomdp_r=pomdp_r)
 
-    factory_kwargs = dict(n_agents=n_agents, max_steps=max_steps, obs_prop=obs_props,
+    factory_kwargs = dict(n_agents=n_agents, individual_rewards=individual_rewards,
+                          max_steps=max_steps, obs_prop=obs_props,
                           mv_prop=MovementProperties(**dictionary['movement_props']),
                           dirt_prop=DirtProperties(**dictionary['dirt_props']),
                           record_episodes=False, verbose=False, **dictionary['factory_props']

environments/factory/base/base_factory.py

@@ -15,12 +15,11 @@ from environments.helpers import Constants as c, Constants
 from environments import helpers as h
 from environments.factory.base.objects import Agent, Tile, Action
 from environments.factory.base.registers import Actions, Entities, Agents, Doors, FloorTiles, WallTiles, PlaceHolders
-from environments.utility_classes import MovementProperties, ObservationProperties
+from environments.utility_classes import MovementProperties, ObservationProperties, MarlFrameStack
 from environments.utility_classes import AgentRenderOptions as a_obs
 
 import simplejson
 
-
 REC_TAC = 'rec_'
 
 
@@ -57,7 +56,7 @@ class BaseFactory(gym.Env):
 
     def __enter__(self):
         return self if self.obs_prop.frames_to_stack == 0 else \
-            FrameStack(self, self.obs_prop.frames_to_stack)
+            MarlFrameStack(FrameStack(self, self.obs_prop.frames_to_stack))
 
     def __exit__(self, exc_type, exc_val, exc_tb):
         self.close()

environments/utility_classes.py

@@ -1,4 +1,6 @@
 from typing import NamedTuple, Union
+import gym
+from gym.wrappers.frame_stack import FrameStack
 
 
 class AgentRenderOptions(object):
@@ -21,3 +23,14 @@ class ObservationProperties(NamedTuple):
     cast_shadows = True
     frames_to_stack: int = 0
     pomdp_r: int = 0
+
+
+class MarlFrameStack(gym.ObservationWrapper):
+    def __init__(self, env):
+        super().__init__(env)
+
+    def observation(self, observation):
+        if isinstance(self.env, FrameStack) and self.env.unwrapped.n_agents > 1:
+            return observation[0:].swapaxes(0, 1)
+        return observation
+

studies/sat_mad.py

@@ -9,15 +9,22 @@ from pathlib import Path
 import numpy as np
 from tqdm import tqdm
 import time
-from algorithms.utils import add_env_props, load_yaml_file, CombineActionsAgent
+from algorithms.utils import (
+    add_env_props,
+    load_yaml_file,
+    CombineActionsAgent,
+    AutoResetGymMultiAgent,
+    access_str,
+    AGENT_PREFIX, REWARD, CUMU_REWARD, OBS, SEP
+)
 
 
 class A2CAgent(TAgent):
-    def __init__(self, observation_size, hidden_size, n_actions, agent_id=-1, marl=False):
+    def __init__(self, observation_size, hidden_size, n_actions, agent_id):
         super().__init__()
         observation_size = np.prod(observation_size)
+        print(observation_size)
         self.agent_id = agent_id
-        self.marl = marl
         self.model = nn.Sequential(
             nn.Flatten(),
             nn.Linear(observation_size, hidden_size),
@@ -31,10 +38,7 @@ class A2CAgent(TAgent):
         self.critic_head = nn.Linear(hidden_size, 1)
 
     def get_obs(self, t):
-        observation = self.get(("env/env_obs", t))
-        if self.marl:
-            observation = observation.permute(2, 0, 1, 3, 4, 5)
-            observation = observation[self.agent_id]
+        observation = self.get((f'env/{access_str(self.agent_id, OBS)}', t))
         return observation
 
     def forward(self, t, stochastic, **kwargs):
@@ -47,17 +51,16 @@ class A2CAgent(TAgent):
             action = torch.distributions.Categorical(probs).sample()
         else:
             action = probs.argmax(1)
-        agent_str = f'agent{self.agent_id}_'
-        self.set((f'{agent_str}action', t), action)
-        self.set((f'{agent_str}action_probs', t), probs)
-        self.set((f'{agent_str}critic', t), critic)
+        self.set((f'{access_str(self.agent_id, "action")}', t), action)
+        self.set((f'{access_str(self.agent_id, "action_probs")}', t), probs)
+        self.set((f'{access_str(self.agent_id, "critic")}', t), critic)
 
 
 if __name__ == '__main__':
     # Setup workspace
     uid = time.time()
     workspace = Workspace()
-    n_agents = 1
+    n_agents = 2
 
     # load config
     cfg = load_yaml_file(Path(__file__).parent / 'sat_mad.yaml')
@@ -65,15 +68,14 @@ if __name__ == '__main__':
     cfg['env'].update({'n_agents': n_agents})
 
     # instantiate agent and env
-    env_agent = AutoResetGymAgent(
+    env_agent = AutoResetGymMultiAgent(
         get_class(cfg['env']),
         get_arguments(cfg['env']),
         n_envs=1
     )
 
     a2c_agents = [instantiate_class({**cfg['agent'],
-                                     'agent_id': agent_id,
-                                     'marl':     n_agents > 1})
+                                     'agent_id': agent_id})
                   for agent_id in range(n_agents)]
 
     # combine agents
@@ -99,11 +101,13 @@ if __name__ == '__main__':
 
             for agent_id in range(n_agents):
                 critic, done, action_probs, reward, action = workspace[
-                    f"agent{agent_id}_critic", "env/done",
-                    f'agent{agent_id}_action_probs', "env/reward",
-                    f"agent{agent_id}_action"
+                    access_str(agent_id, 'critic'),
+                    "env/done",
+                    access_str(agent_id, 'action_probs'),
+                    access_str(agent_id, 'reward', 'env/'),
+                    access_str(agent_id, 'action')
                 ]
-                td = gae(critic, reward, done, 0.99, 0.3)
+                td = gae(critic, reward, done, 0.98, 0.25)
                 td_error = td ** 2
                 critic_loss = td_error.mean()
                 entropy_loss = Categorical(action_probs).entropy().mean()
@@ -118,16 +122,18 @@ if __name__ == '__main__':
                 optimizer = optimizers[agent_id]
                 optimizer.zero_grad()
                 loss.backward()
-                #torch.nn.utils.clip_grad_norm_(a2c_agents[agent_id].parameters(), 2)
+                #torch.nn.utils.clip_grad_norm_(a2c_agents[agent_id].parameters(), .5)
                 optimizer.step()
 
                 # Compute the cumulated reward on final_state
-                creward = workspace["env/cumulated_reward"]
+                rews = ''
+                for agent_i in range(n_agents):
+                    creward = workspace['env/'+access_str(agent_i, CUMU_REWARD)]
                     creward = creward[done]
                     if creward.size()[0] > 0:
-                    cum_r = creward.mean().item()
-                    if cum_r > best:
-                    #    torch.save(a2c_agent.state_dict(), Path(__file__).parent / f'agent_{uid}.pt')
-                        best = cum_r
-                    pbar.set_description(f"Cum. r: {cum_r:.2f}, Best r. so far: {best:.2f}", refresh=True)
+                        rews += f'{AGENT_PREFIX}{agent_i}: {creward.mean().item():.2f}  |  '
+                        """if cum_r > best:
+                            torch.save(a2c_agent.state_dict(), Path(__file__).parent / f'agent_{uid}.pt')
+                            best = cum_r"""
+                        pbar.set_description(rews, refresh=True)
 

studies/sat_mad.yaml

@@ -11,15 +11,16 @@ env:
   pomdp_r:            2
   max_steps:          400
   stack_n_frames:     3
+  individual_rewards: True
 
 algorithm:
   max_epochs:             1000000
   n_envs:                 1
-  n_timesteps:            16
+  n_timesteps:            10
   discount_factor:        0.99
   entropy_coef:           0.01
   critic_coef:            1.0
-  gae:                    0.3
+  gae:                    0.25
   optimizer:
     classname:            torch.optim.Adam
     lr:                   0.0003