evolutionary-policy-optimization 0.2.11__tar.gz → 0.2.14__tar.gz

{evolutionary_policy_optimization-0.2.11 → evolutionary_policy_optimization-0.2.14}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: evolutionary-policy-optimization
-Version: 0.2.11
+Version: 0.2.14
 Summary: EPO - Pytorch
 Project-URL: Homepage, https://pypi.org/project/evolutionary-policy-optimization/
 Project-URL: Repository, https://github.com/lucidrains/evolutionary-policy-optimization
@@ -87,10 +87,10 @@ latent_pool = LatentGenePool(
     dim_latent = 32,
 )
 
-state = torch.randn(1, 512)
+state = torch.randn(1, 32)
 
-actor = Actor(512, dim_hiddens = (256, 128), num_actions = 4, dim_latent = 32)
-critic = Critic(512, dim_hiddens = (256, 128, 64), dim_latent = 32)
+actor = Actor(dim_state = 32, dim = 256, mlp_depth = 2, num_actions = 4, dim_latent = 32)
+critic = Critic(dim_state = 32, dim = 256, mlp_depth = 3, dim_latent = 32)
 
 latent = latent_pool(latent_id = 2)
 
@@ -289,4 +289,15 @@ That's it
 }
 ```
 
+```bibtex
+@article{Jiang2022GeneralIR,
+    title     = {General intelligence requires rethinking exploration},
+    author    = {Minqi Jiang and Tim Rocktaschel and Edward Grefenstette},
+    journal   = {Royal Society Open Science},
+    year      = {2022},
+    volume    = {10},
+    url       = {https://api.semanticscholar.org/CorpusID:253523156}
+}
+```
+
 *Evolution is cleverer than you are.* - Leslie Orgel

{evolutionary_policy_optimization-0.2.11 → evolutionary_policy_optimization-0.2.14}/README.md

@@ -30,10 +30,10 @@ latent_pool = LatentGenePool(
     dim_latent = 32,
 )
 
-state = torch.randn(1, 512)
+state = torch.randn(1, 32)
 
-actor = Actor(512, dim_hiddens = (256, 128), num_actions = 4, dim_latent = 32)
-critic = Critic(512, dim_hiddens = (256, 128, 64), dim_latent = 32)
+actor = Actor(dim_state = 32, dim = 256, mlp_depth = 2, num_actions = 4, dim_latent = 32)
+critic = Critic(dim_state = 32, dim = 256, mlp_depth = 3, dim_latent = 32)
 
 latent = latent_pool(latent_id = 2)
 
@@ -232,4 +232,15 @@ That's it
 }
 ```
 
+```bibtex
+@article{Jiang2022GeneralIR,
+    title     = {General intelligence requires rethinking exploration},
+    author    = {Minqi Jiang and Tim Rocktaschel and Edward Grefenstette},
+    journal   = {Royal Society Open Science},
+    year      = {2022},
+    volume    = {10},
+    url       = {https://api.semanticscholar.org/CorpusID:253523156}
+}
+```
+
 *Evolution is cleverer than you are.* - Leslie Orgel

{evolutionary_policy_optimization-0.2.11 → evolutionary_policy_optimization-0.2.14}/evolutionary_policy_optimization/epo.py

@@ -589,9 +589,9 @@ class Critic(Module):
         use_regression = False,
         state_norm: StateNorm | None = None,
         hl_gauss_loss_kwargs: dict = dict(
-            min_value = -100.,
-            max_value = 100.,
-            num_bins = 200
+            min_value = -10.,
+            max_value = 10.,
+            num_bins = 250
         )
     ):
         super().__init__()
@@ -644,11 +644,11 @@ class Critic(Module):
         loss_fn = partial(self.loss_fn, reduction = 'none')
 
         if use_improved:
-            clipped_target = target.clamp(-eps_clip, eps_clip)
-
             old_values_lo = old_values - eps_clip
             old_values_hi = old_values + eps_clip
 
+            clipped_target = target.clamp(old_values_lo, old_values_hi)
+
             def is_between(lo, hi):
                 return (lo < value) & (value < hi)
 

{evolutionary_policy_optimization-0.2.11 → evolutionary_policy_optimization-0.2.14}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "evolutionary-policy-optimization"
-version = "0.2.11"
+version = "0.2.14"
 description = "EPO - Pytorch"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }