Merge branch 'johan/devbranch' into johan/micromacro

Author: Johanmkr

pyproject.toml

@@ -21,6 +21,7 @@ dependencies = [
     "sphinx-rtd-theme>=3.0.2",
     "torch>=2.6.0",
     "torchvision>=0.21.0",
+    "tqdm>=4.67.1",
 ]
 [tool.isort]
 profile = "black"

utils/metrics/precision.py

@@ -13,49 +13,95 @@ class Precision(nn.Module):
     ----------
     num_classes : int
         Number of classes in the dataset.
-    use_mean : bool
-        Whether to calculate precision as a mean of precisions or as a brute function of true positives and false positives.
+    micro_averaging : bool
+        Wheter to compute the micro or macro precision (default False)
     """
 
-    def __init__(self, num_classes: int, use_mean: bool = True):
+    def __init__(self, num_classes: int, micro_averaging: bool = False):
         super().__init__()
 
         self.num_classes = num_classes
-        self.use_mean = use_mean
+        self._micro_averaging = micro_averaging
 
     def forward(self, y_true: torch.tensor, y_pred: torch.tensor) -> torch.tensor:
-        """Calculates the precision score given number of classes and the true and predicted labels.
+        """Compute precision of model
 
         Parameters
         ----------
         y_true : torch.tensor
-            true labels
+            True labels
         y_pred : torch.tensor
-            predicted labels
+            Predicted labels
 
         Returns
         -------
         torch.tensor
-            precision score
+            Precision score
+        """
+        return (
+            self._micro_avg_precision(y_true, y_pred)
+            if self.micro_averaging
+            else self._macro_avg_precision(y_true, y_pred)
+        )
+
+    def _micro_avg_precision(
+        self, y_true: torch.tensor, y_pred: torch.tensor
+    ) -> torch.tensor:
+        """Compute micro-average precision by first calculating true/false positive across all classes and then find the precision.
+
+        Parameters
+        ----------
+        y_true : torch.tensor
+            True labels
+        y_pred : torch.tensor
+            Predicted labels
+
+        Returns
+        -------
+        torch.tensor
+            Micro-averaged precision
         """
-        # One-hot encode the target tensor
         true_oh = torch.zeros(y_true.size(0), self.num_classes).scatter_(
             1, y_true.unsqueeze(1), 1
         )
         pred_oh = torch.zeros(y_pred.size(0), self.num_classes).scatter_(
             1, y_pred.unsqueeze(1), 1
         )
+        tp = torch.sum(true_oh * pred_oh)
+        fp = torch.sum(~true_oh[pred_oh.bool()].bool())
 
-        if self.use_mean:
-            tp = torch.sum(true_oh * pred_oh, 0)
-            fp = torch.sum(~true_oh.bool() * pred_oh, 0)
+        return torch.nanmean(tp / (tp + fp))
+
+    def _macro_avg_precision(
+        self, y_true: torch.tensor, y_pred: torch.tensor
+    ) -> torch.tensor:
+        """Compute macro-average precision by finding true/false positives of each class separately then averaging across all classes.
 
-        else:
-            tp = torch.sum(true_oh * pred_oh)
-            fp = torch.sum(~true_oh[pred_oh.bool()].bool())
+        Parameters
+        ----------
+        y_true : torch.tensor
+            True labels
+        y_pred : torch.tensor
+            Predicted labels
+
+        Returns
+        -------
+        torch.tensor
+            Macro-averaged precision
+        """
+        true_oh = torch.zeros(y_true.size(0), self.num_classes).scatter_(
+            1, y_true.unsqueeze(1), 1
+        )
+        pred_oh = torch.zeros(y_pred.size(0), self.num_classes).scatter_(
+            1, y_pred.unsqueeze(1), 1
+        )
+        tp = torch.sum(true_oh * pred_oh, 0)
+        fp = torch.sum(~true_oh.bool() * pred_oh, 0)
 
         return torch.nanmean(tp / (tp + fp))
 
 
 if __name__ == "__main__":
-    pass
+    print(
+        "Congratulations, you succesfully ran the Precision metric class. You should be proud of this marvelous achievement!"
+    )