optionally keep track of returns statistics and normalize with them before advantage

dreamer4/dreamer4.py

@@ -1817,6 +1817,8 @@ class DynamicsWorldModel(Module):
         continuous_action_loss_weight: float | list[float] = 1.,
         num_latent_genes = 0,                       # for carrying out evolution within the dreams https://web3.arxiv.org/abs/2503.19037
         num_residual_streams = 1,
+        keep_reward_ema_stats = False,
+        reward_ema_decay = 0.998,
         gae_discount_factor = 0.997,
         gae_lambda = 0.95,
         ppo_eps_clip = 0.2,
@@ -2022,6 +2024,14 @@ class DynamicsWorldModel(Module):
         self.value_clip = value_clip
         self.policy_entropy_weight = value_clip
 
+        # rewards related
+
+        self.keep_reward_ema_stats = keep_reward_ema_stats
+        self.reward_ema_decay = reward_ema_decay
+
+        self.register_buffer('ema_returns_mean', tensor(0.))
+        self.register_buffer('ema_returns_var', tensor(1.))
+
         # loss related
 
         self.flow_loss_normalizer = LossNormalizer(1)
@@ -2267,11 +2277,32 @@ class DynamicsWorldModel(Module):
 
         world_model_forward_context = torch.no_grad if only_learn_policy_value_heads else nullcontext
 
+        # maybe keep track returns statistics and normalize returns and values before calculating advantage, as done in dreamer v3
+
+        if self.keep_reward_ema_stats:
+            ema_returns_mean, ema_returns_var = self.ema_returns_mean, self.ema_returns_var
+
+            decay = 1. - self.reward_ema_decay
+
+            # todo - handle distributed
+
+            returns_mean, returns_var = returns.mean(), returns.var()
+
+            ema_returns_mean.lerp_(returns_mean, decay)
+            ema_returns_var.lerp_(returns_var, decay)
+
+            ema_returns_std = ema_returns_var.clamp(min = 1e-5).sqrt()
+
+            normed_returns = (returns - ema_returns_mean) / ema_returns_std
+            normed_old_values = (old_values - ema_returns_mean) / ema_returns_std
+
+            advantage = normed_returns - normed_old_values
+        else:
+            advantage = returns - old_values
+
         # apparently they just use the sign of the advantage
         # https://arxiv.org/abs/2410.04166v1
 
-        advantage = returns - old_values
-
         if use_signed_advantage:
             advantage = advantage.sign()
         else:

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "dreamer4"
-version = "0.0.80"
+version = "0.0.81"
 description = "Dreamer 4"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }