Compatibility update, release 0.5.3

trackmania-rl · Oct 10, 2023 · e400412 · e400412
1 parent e325488
commit e400412
Show file tree

Hide file tree

Showing 6 changed files with 22 additions and 21 deletions.
diff --git a/readme/tuto_library.md b/readme/tuto_library.md
@@ -1161,11 +1161,11 @@ Note that `train()` returns a python dictionary in which you can store the metri
                 p_targ.data.mul_(self.polyak)
                 p_targ.data.add_((1 - self.polyak) * p.data)
         ret_dict = dict(
-            loss_actor=loss_pi.detach(),
-            loss_critic=loss_q.detach(),
+            loss_actor=loss_pi.detach().item(),
+            loss_critic=loss_q.detach().item(),
         )
         if self.learn_entropy_coef:
-            ret_dict["loss_entropy_coef"] = loss_alpha.detach()
+            ret_dict["loss_entropy_coef"] = loss_alpha.detach().item()
             ret_dict["entropy_coef"] = alpha_t.item()
         return ret_dict  # dictionary of metrics to be logged
 ```

diff --git a/setup.py b/setup.py
@@ -100,7 +100,7 @@ def url_retrieve(url: str, outfile: Path, overwrite: bool = False):
 
 install_req = [
     'numpy',
-    'torch',
+    'torch>=2.0.0',
     'pandas',
     'gymnasium',
     'rtgym>=0.9',
@@ -111,7 +111,8 @@ def url_retrieve(url: str, outfile: Path, overwrite: bool = False):
     'keyboard',
     'pyautogui',
     'pyinstrument',
-    'tlspyo>=0.2.5'
+    'tlspyo>=0.2.5',
+    'chardet'  # requests dependency
 ]
 
 if platform.system() == "Windows":
@@ -127,13 +128,13 @@ def url_retrieve(url: str, outfile: Path, overwrite: bool = False):
 
 setup(
     name='tmrl',
-    version='0.5.2',
+    version='0.5.3',
     description='Network-based framework for real-time robot learning',
     long_description=README,
     long_description_content_type='text/markdown',
     keywords='reinforcement learning, robot learning, trackmania, self driving, roborace',
     url='https://github.com/trackmania-rl/tmrl',
-    download_url='https://github.com/trackmania-rl/tmrl/archive/refs/tags/v0.5.2.tar.gz',
+    download_url='https://github.com/trackmania-rl/tmrl/archive/refs/tags/v0.5.3.tar.gz',
     author='Yann Bouteiller, Edouard Geze',
     author_email='[email protected], [email protected]',
     license='MIT',

diff --git a/tmrl/custom/custom_algorithms.py b/tmrl/custom/custom_algorithms.py
@@ -157,8 +157,8 @@ def train(self, batch):
 
             if not cfg.DEBUG_MODE:
                 ret_dict = dict(
-                    loss_actor=loss_pi.detach(),
-                    loss_critic=loss_q.detach(),
+                    loss_actor=loss_pi.detach().item(),
+                    loss_critic=loss_q.detach().item(),
                 )
             else:
                 q1_o2_a2 = self.model.q1(o2, a2)
@@ -182,8 +182,8 @@ def train(self, batch):
                 diff_q2_backup_r = (q2 - backup + r).detach()
 
                 ret_dict = dict(
-                    loss_actor=loss_pi.detach(),
-                    loss_critic=loss_q.detach(),
+                    loss_actor=loss_pi.detach().item(),
+                    loss_critic=loss_q.detach().item(),
                     # debug:
                     debug_log_pi=logp_pi.detach().mean(),
                     debug_log_pi_std=logp_pi.detach().std(),
@@ -248,7 +248,7 @@ def train(self, batch):
                 )
 
         if self.learn_entropy_coef:
-            ret_dict["loss_entropy_coef"] = loss_alpha.detach()
+            ret_dict["loss_entropy_coef"] = loss_alpha.detach().item()
             ret_dict["entropy_coef"] = alpha_t.item()
 
         return ret_dict
@@ -375,12 +375,12 @@ def train(self, batch):
         if update_policy:
             self.loss_pi = loss_pi.detach()
         ret_dict = dict(
-            loss_actor=self.loss_pi,
-            loss_critic=loss_q.detach(),
+            loss_actor=self.loss_pi.detach().item(),
+            loss_critic=loss_q.detach().item(),
         )
 
         if self.learn_entropy_coef:
-            ret_dict["loss_entropy_coef"] = loss_alpha.detach()
+            ret_dict["loss_entropy_coef"] = loss_alpha.detach().item()
             ret_dict["entropy_coef"] = alpha_t.item()
 
         return ret_dict
diff --git a/tmrl/custom/utils/nn.py b/tmrl/custom/utils/nn.py
@@ -41,7 +41,7 @@ def copy_shared(model_a):
     for key in sda:
         a, b = sda[key], sdb[key]
         b.data = a.data  # strangely this will not make a.data and b.data the same object but their underlying data_ptr will be the same
-        assert b.storage().data_ptr() == a.storage().data_ptr()
+        assert b.untyped_storage().data_ptr() == a.untyped_storage().data_ptr()
     return model_b
 
 

diff --git a/tmrl/tuto/competition/custom_actor_module.py b/tmrl/tuto/competition/custom_actor_module.py
@@ -781,8 +781,8 @@ def train(self, batch):
 
         # TMRL enables us to log training metrics to wandb:
         ret_dict = dict(
-            loss_actor=loss_pi.detach(),
-            loss_critic=loss_q.detach(),
+            loss_actor=loss_pi.detach().item(),
+            loss_critic=loss_q.detach().item(),
         )
         return ret_dict
 

diff --git a/tmrl/tuto/tuto.py b/tmrl/tuto/tuto.py
@@ -551,11 +551,11 @@ def train(self, batch):
                 p_targ.data.mul_(self.polyak)
                 p_targ.data.add_((1 - self.polyak) * p.data)
         ret_dict = dict(
-            loss_actor=loss_pi.detach(),
-            loss_critic=loss_q.detach(),
+            loss_actor=loss_pi.detach().item(),
+            loss_critic=loss_q.detach().item(),
         )
         if self.learn_entropy_coef:
-            ret_dict["loss_entropy_coef"] = loss_alpha.detach()
+            ret_dict["loss_entropy_coef"] = loss_alpha.detach().item()
             ret_dict["entropy_coef"] = alpha_t.item()
         return ret_dict