evolutionary-policy-optimization 0.2.10__py3-none-any.whl → 0.2.12__py3-none-any.whl

evolutionary_policy_optimization/distributed.py

@@ -61,6 +61,8 @@ def has_only_one_value(t):
     return (t == t[0]).all()
 
 def all_gather_variable_dim(t, dim = 0, sizes = None):
+    device = t.device
+
     if not exists(sizes):
         sizes = gather_sizes(t, dim = dim)
 
@@ -77,7 +79,7 @@ def all_gather_variable_dim(t, dim = 0, sizes = None):
     gathered_tensors = torch.cat(gathered_tensors, dim = dim)
     seq = torch.arange(max_size, device = device)
 
-    mask = einx.less('j i -> (i j)', seq, sizes)
+    mask = einx.less('j, i -> (i j)', seq, sizes)
     seq = torch.arange(mask.shape[-1], device = device)
     indices = seq[mask]
 

evolutionary_policy_optimization/epo.py

@@ -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-0.2.10.dist-info → evolutionary_policy_optimization-0.2.12.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: evolutionary-policy-optimization
-Version: 0.2.10
+Version: 0.2.12
 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-0.2.10.dist-info → evolutionary_policy_optimization-0.2.12.dist-info}/RECORD

@@ -1,10 +1,10 @@
 evolutionary_policy_optimization/__init__.py,sha256=NyiYDYU7DlpmOTM7xiBQET3r1WwX0ebrgMCBLSQrW3c,288
-evolutionary_policy_optimization/distributed.py,sha256=MxyxqxANAuOm8GYb0Yu09EHd_aVLhK2uwgrfuVWciPU,2342
+evolutionary_policy_optimization/distributed.py,sha256=clN8Bvhb6JIQy2F8FTF312B0RN3iYoPkKjZreBtAwks,2366
 evolutionary_policy_optimization/env_wrappers.py,sha256=bDL06o9_b1iW6k3fw2xifnOnYlzs643tdW6Yv2gsIdw,803
-evolutionary_policy_optimization/epo.py,sha256=adUHDtgrXnSOoPLnbPy6xxGl6QLYxbN1mB_sl2KPwgI,53135
+evolutionary_policy_optimization/epo.py,sha256=9zOGucA3jWODOdtk9lBfUAaScOnbVfTLqYV8_x1BhKk,53144
 evolutionary_policy_optimization/experimental.py,sha256=QZG0__wwFqHN_LJK7e-mHxlIL1mwjlvG6o6bcOpeAKg,6166
 evolutionary_policy_optimization/mock_env.py,sha256=TLyyRm6tOD0Kdn9QqJJQriaSnsR-YmNQHo4OohmZFG4,1410
-evolutionary_policy_optimization-0.2.10.dist-info/METADATA,sha256=dLg2z2BFihNUd0dirhDcabw9yNj0V_2v42Tk_Pg-EwY,9601
-evolutionary_policy_optimization-0.2.10.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-evolutionary_policy_optimization-0.2.10.dist-info/licenses/LICENSE,sha256=1yCiA9b5nhslTavxPjsQAO-wpOnwJR9-l8LTVi7GJuk,1066
-evolutionary_policy_optimization-0.2.10.dist-info/RECORD,,
+evolutionary_policy_optimization-0.2.12.dist-info/METADATA,sha256=tZIKqUx-akzbJZWd0h0hM9F8FWbq6SjpJ7Vqstk4dNo,9972
+evolutionary_policy_optimization-0.2.12.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+evolutionary_policy_optimization-0.2.12.dist-info/licenses/LICENSE,sha256=1yCiA9b5nhslTavxPjsQAO-wpOnwJR9-l8LTVi7GJuk,1066
+evolutionary_policy_optimization-0.2.12.dist-info/RECORD,,