homa 0.3.11__tar.gz → 0.3.12__tar.gz

{homa-0.3.11 → homa-0.3.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: homa
-Version: 0.3.11
+Version: 0.3.12
 Summary: A curated list of machine learning and deep learning helpers.
 Author-email: Taha Shieenavaz <tahashieenavaz@gmail.com>
 Requires-Python: >=3.7

{homa-0.3.11 → homa-0.3.12}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "homa"
-version = "0.3.11"
+version = "0.3.12"
 description = "A curated list of machine learning and deep learning helpers."
 authors = [
   { name="Taha Shieenavaz", email="tahashieenavaz@gmail.com" },

{homa-0.3.11 → homa-0.3.12}/src/homa/rl/buffers/Buffer.py

@@ -1,10 +1,12 @@
+from collections import deque
+from typing import Type
 from .concerns import ResetsCollection, HasRecordAlternatives
 
 
 class Buffer(ResetsCollection, HasRecordAlternatives):
     def __init__(self, capacity: int):
         self.capacity: int = capacity
-        self.reset()
+        self.collection: Type[deque] = deque(maxlen=self.capacity)
 
     @property
     def size(self):

{homa-0.3.11 → homa-0.3.12}/src/homa/rl/buffers/SoftActorCriticBuffer.py

@@ -35,7 +35,7 @@ class SoftActorCriticBuffer(Buffer):
 
         if as_tensor:
             states = torch.from_numpy(states).float()
-            actions = torch.from_numpy(actions).long()
+            actions = torch.from_numpy(actions).float()
             rewards = torch.from_numpy(rewards).float()
             next_states = torch.from_numpy(next_states).float()
             terminations = torch.from_numpy(terminations).float()

{homa-0.3.11 → homa-0.3.12}/src/homa/rl/sac/SoftCritic.py

@@ -1,6 +1,5 @@
 import torch
 from torch.nn.functional import mse_loss as mse
-from typing import Type
 from .modules import DualSoftCriticModule
 from .SoftActor import SoftActor
 
@@ -31,6 +30,10 @@ class SoftCritic:
             hidden_dimension=hidden_dimension,
             action_dimension=action_dimension,
         )
+
+        # copy source to target when initiated
+        self.target.load_state_dict(self.network.state_dict())
+
         self.optimizer = torch.optim.AdamW(
             self.network.parameters(), lr=lr, weight_decay=weight_decay
         )
@@ -65,7 +68,7 @@ class SoftCritic:
         next_states: torch.Tensor,
         actor: torch.nn.Module,
     ):
-        q_alpha, q_beta = self.target(states, actions)
+        q_alpha, q_beta = self.network(states, actions)
         target = self.calculate_target(
             rewards=rewards,
             terminations=terminations,
@@ -82,16 +85,15 @@ class SoftCritic:
         next_states: torch.Tensor,
         actor: SoftActor,
     ):
+        termination_mask = 1 - terminations
         next_actions, next_probabilities = actor.sample(next_states)
         q_alpha, q_beta = self.target(next_states, next_actions)
         q = torch.min(q_alpha, q_beta)
-        termination_mask = 1 - terminations
         entropy_q = q - self.alpha * next_probabilities * termination_mask
         return rewards + self.gamma * entropy_q
 
-    def soft_update(
-        self, network: Type[torch.nn.Module], target: Type[torch.nn.Module]
-    ):
+    @torch.no_grad()
+    def soft_update(self, network: torch.nn.Module, target: torch.nn.Module):
         for s, t in zip(network.parameters(), target.parameters()):
             t.data.copy_(self.tau * s.data + (1 - self.tau) * t.data)
 

{homa-0.3.11 → homa-0.3.12}/src/homa/rl/sac/modules/SoftActorModule.py

@@ -30,6 +30,6 @@ class SoftActorModule(torch.nn.Module):
     def forward(self, state: torch.Tensor):
         features = self.phi(state)
         mean = self.mu(features)
-        std = self.mu(features)
+        std = self.xi(features)
         std = std.clamp(self.min_std, self.max_std)
         return mean, std

{homa-0.3.11 → homa-0.3.12}/src/homa.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: homa
-Version: 0.3.11
+Version: 0.3.12
 Summary: A curated list of machine learning and deep learning helpers.
 Author-email: Taha Shieenavaz <tahashieenavaz@gmail.com>
 Requires-Python: >=3.7