Fix eta_min parameter in lr_scheduler.

gyshgx868 · Apr 13, 2020 · 3b9bda5 · 3b9bda5
1 parent 920ba94
commit 3b9bda5
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/graph_ter_cls/runner/runner.py b/graph_ter_cls/runner/runner.py
@@ -154,7 +154,7 @@ def load_scheduler(self):
         if self.optimizer is None:
             return
         self.scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
-            self.optimizer, self.args.num_epochs, eta_min=self.args.lr
+            self.optimizer, self.args.num_epochs, eta_min=self.args.lr / 100.0
         )
 
     @abstractmethod

diff --git a/graph_ter_seg/runner/runner.py b/graph_ter_seg/runner/runner.py
@@ -148,7 +148,7 @@ def load_scheduler(self):
         if self.model['train'] is None:
             return
         self.scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
-            self.optimizer, self.args.num_epochs, eta_min=self.args.lr
+            self.optimizer, self.args.num_epochs, eta_min=self.args.lr / 100.0
         )
 
     @abstractmethod