Use balanced accuracy as base metric

w4k2 · Jan 19, 2024 · 860e913 · 860e913
1 parent d26c542
commit 860e913
Show file tree

Hide file tree

Showing 7 changed files with 15 additions and 13 deletions.
diff --git a/README.md b/README.md
@@ -171,7 +171,7 @@ ax.plot(results.T, label=['Inner', 'Outer', 'Halfpoint', 'Overall'])
 ax.legend()
 ax.grid(ls=':')
 ax.set_xlabel('epochs')
-ax.set_ylabel('Weighted accurracy')
+ax.set_ylabel('Balanced accurracy')
 ax.set_xlim(0,epochs)
 ```
 

diff --git a/docs/source/quickstart.rst b/docs/source/quickstart.rst
@@ -156,7 +156,7 @@ The results of the single processing can be visualized using `matplotlib` librar
   ax.legend()
   ax.grid(ls=':')
   ax.set_xlabel('epochs')
-  ax.set_ylabel('Weighted accurracy')
+  ax.set_ylabel('Balanced accurracy')
   ax.set_xlim(0,epochs)
 
 .. image:: _static/example.png

diff --git a/example.png b/example.png
diff --git a/torchosr/_version.py b/torchosr/_version.py
@@ -1,3 +1,3 @@
 """``torchosr``"""
 
-__version__ = "0.1.2"
+__version__ = "0.1.3"
diff --git a/torchosr/models/GSL.py b/torchosr/models/GSL.py
@@ -85,14 +85,15 @@ def test(self, dataloader, loss_fn, conf=False):
         outer_targets = []
         overall_targets = []
 
-        # Define metric for open set detection
+        # Define metrics for open set detection
         try:
-            inner_metric = Accuracy(task="multiclass", num_classes=self.n_known, average='weighted')
-            overall_metric = Accuracy(task="multiclass", num_classes=self.n_known+1, average='weighted')
+            inner_metric = Accuracy(task="multiclass", num_classes=self.n_known, average='macro')
+            overall_metric = Accuracy(task="multiclass", num_classes=self.n_known+1, average='macro')
         except:
             inner_metric = None
             overall_metric = None
-        outer_metric = Accuracy(task='binary', average='weighted')
+
+        outer_metric = Accuracy(task='multiclass', num_classes=2, average='macro')
 
         with torch.no_grad():
             for X, y in dataloader:

diff --git a/torchosr/models/Openmax.py b/torchosr/models/Openmax.py
@@ -106,13 +106,13 @@ def test(self, dataloader, loss_fn, conf=False):
 
         # Define metrics for open set detection
         try:
-            inner_metric = Accuracy(task="multiclass", num_classes=self.n_known, average='weighted')
-            overall_metric = Accuracy(task="multiclass", num_classes=self.n_known+1, average='weighted')
+            inner_metric = Accuracy(task="multiclass", num_classes=self.n_known, average='macro')
+            overall_metric = Accuracy(task="multiclass", num_classes=self.n_known+1, average='macro')
         except:
             inner_metric = None
             overall_metric = None
 
-        outer_metric = Accuracy(task='binary', average='weighted')
+        outer_metric = Accuracy(task='multiclass', num_classes=2, average='macro')
 
         with torch.no_grad():
             for X, y in dataloader:

diff --git a/torchosr/models/TSoftmax.py b/torchosr/models/TSoftmax.py
@@ -38,11 +38,12 @@ def test(self, dataloader, loss_fn, conf=False):
 
         # Define metric for open set detection
         try:
-            inner_metric = Accuracy(task="multiclass", num_classes=self.n_known, average='weighted')
-            overall_metric = Accuracy(task="multiclass", num_classes=self.n_known+1, average='weighted')
+            inner_metric = Accuracy(task="multiclass", num_classes=self.n_known, average='macro')
+            overall_metric = Accuracy(task="multiclass", num_classes=self.n_known+1, average='macro')
         except:
             inner_metric = None
-        outer_metric = Accuracy(task='binary', average='weighted')
+
+        outer_metric = Accuracy(task='multiclass', num_classes=2, average='macro')
 
         with torch.no_grad():
             for X, y in dataloader: