Fix rendering

- support predict()
- move discount to OptionsEnv
- fix RenderOptions

scratch/etienne/intersimple/gail/options2.py

@@ -15,7 +15,7 @@ def imitation_discriminator(discriminator):
 
 class OptionsEnv(gym.Wrapper):
 
-    def __init__(self, env, options, discriminator, ll_buffer_capacity, *args, **kwargs):
+    def __init__(self, env, options, discriminator, discount, ll_buffer_capacity, *args, **kwargs):
         super().__init__(env, *args, **kwargs)
 
         self.options = options
@@ -27,6 +27,7 @@ class OptionsEnv(gym.Wrapper):
         })
 
         self.discriminator = discriminator
+        self.discount = discount
         self.ll_buffer_capacity = ll_buffer_capacity
         self.ll_buffer = deque(maxlen=ll_buffer_capacity)
     
@@ -85,11 +86,11 @@ class OptionsEnv(gym.Wrapper):
 
 class RenderOptions(OptionsEnv):
 
-    def __init__(self, options, *args, **kwargs):
-        super().__init__(options, discriminator=lambda s, a, n, d: 0, ll_buffer_capacity=0, *args, **kwargs)
+    def __init__(self, env, options, *args, **kwargs):
+        super().__init__(env, options, discriminator=lambda s, a, n, d: 0, discount=1, ll_buffer_capacity=0, *args, **kwargs)
 
-    def _ll_step(self):
-        out = super()._ll_step()
+    def _ll_step(self, action):
+        out = super()._ll_step(action)
         self.env.render()
         return out
     

scratch/etienne/intersimple/gail_options_image_random_location.py

@@ -34,7 +34,6 @@ def train(
         epochs=200,
     ):
     env = NRasterizedRouteSpeedRandomAgentLocation(**env_settings)
-    env.discount = discount
 
     tempdir = tempfile.TemporaryDirectory(prefix="quickstart")
     tempdir_path = pathlib.Path(tempdir.name)
@@ -51,10 +50,12 @@ def train(
         gen_algo=stable_baselines3.PPO("CnnPolicy", venv), # unused
     )
 
-    options_env = TimeLimit(OptionsEnv(env,
-        discriminator=imitation_discriminator(discriminator),
+    options_env = TimeLimit(OptionsEnv(
+        env,
         options=ALL_OPTIONS,
-        ll_buffer_capacity=expert_batch_size
+        discriminator=imitation_discriminator(discriminator),
+        discount=discount,
+        ll_buffer_capacity=expert_batch_size,
     ), max_episode_steps=10)
     generator = stable_baselines3.PPO(
         OptionsCnnPolicy,
@@ -79,11 +80,15 @@ def train(
     return generator
 
 def video(model_name, env):
-    model = stable_baselines3.PPO.load(model_name)
     env = RenderOptions(env, options=ALL_OPTIONS)
-    for s in env.sample_ll(model):
-        if s['dones']:
-            break
+    model = stable_baselines3.PPO.load(model_name)
+
+    done = False
+    obs = env.reset()
+    while not done:
+        action, _ = model.predict(obs)
+        obs, _, done, _ = env.step(action)
+
     env.close(filestr='render/'+model_name)
 
 def evaluate():

src/policies/options.py

@@ -1,12 +1,13 @@
-import stable_baselines3
+from stable_baselines3.common.policies import ActorCriticPolicy, ActorCriticCnnPolicy
 from torch.distributions import Categorical
 
-class OptionsCnnPolicy(stable_baselines3.common.policies.ActorCriticCnnPolicy):
+class OptionsCnnPolicy(ActorCriticPolicy):
     """
     Class for high-level options policy (generator)
     """
     def __init__(self, observation_space, *args, **kwargs):
-        super().__init__(observation_space['obs'], *args, **kwargs)
+        super().__init__(observation_space, *args, **kwargs)
+        self.cnn_policy = ActorCriticCnnPolicy(observation_space['obs'], *args, **kwargs)
 
     def _prior_distribution(self, s):
         """
@@ -17,9 +18,9 @@ class OptionsCnnPolicy(stable_baselines3.common.policies.ActorCriticCnnPolicy):
             values (torch.tensor): values from critic
             dist (torch.distributions): prior distribution over actions
         """
-        latent_pi, latent_vf, latent_sde = self._get_latent(s)
-        distribution = self._get_action_dist_from_latent(latent_pi, latent_sde)
-        values = self.value_net(latent_vf)
+        latent_pi, latent_vf, latent_sde = self.cnn_policy._get_latent(s)
+        distribution = self.cnn_policy._get_action_dist_from_latent(latent_pi, latent_sde)
+        values = self.cnn_policy.value_net(latent_vf)
         return values, distribution.distribution
 
     def forward(self, obs, eps=1e-6):
@@ -39,6 +40,10 @@ class OptionsCnnPolicy(stable_baselines3.common.policies.ActorCriticCnnPolicy):
         posterior = Categorical((prior.probs + eps) * m)
         ch = posterior.sample()
         return ch, values, posterior.log_prob(ch)
+    
+    def _predict(self, obs, deterministic=False):
+        action, _, _ = self.forward(obs)
+        return action
 
     def evaluate_actions(self, obs, ch, eps=1e-6):
         """