Merge pull request #29 from SFI-Visual-Intelligence/johan/devbranch

Johan/devbranch

Author: c-salomonsen

utils/load_metric.py

@@ -3,7 +3,7 @@
 import numpy as np
 import torch.nn as nn
 
-from .metrics import EntropyPrediction, F1Score
+from .metrics import EntropyPrediction, F1Score, precision
 
 
 class MetricWrapper(nn.Module):
@@ -39,7 +39,7 @@ def _get_metric(self, key):
             case "recall":
                 raise NotImplementedError("Recall score not implemented yet")
             case "precision":
-                raise NotImplementedError("Precision score not implemented yet")
+                return precision()
             case "accuracy":
                 raise NotImplementedError("Accuracy score not implemented yet")
             case _:

utils/metrics/precision.py

@@ -0,0 +1,105 @@
+import torch
+import torch.nn as nn
+
+USE_MEAN = True
+
+# Precision = TP / (TP + FP)
+
+
+class Precision(nn.Module):
+    """Metric module for precision. Can calculate precision both as a mean of precisions or as brute function of true positives and false positives. This is for now controller with the USE_MEAN macro.
+
+    Parameters
+    ----------
+    num_classes : int
+        Number of classes in the dataset.
+    """
+
+    def __init__(self, num_classes):
+        super().__init__()
+
+        self.num_classes = num_classes
+
+    def forward(self, y_true, y_pred):
+        """Calculates the precision score given number of classes and the true and predicted labels.
+
+        Parameters
+        ----------
+        y_true : torch.tensor
+            true labels
+        y_pred : torch.tensor
+            predicted labels
+
+        Returns
+        -------
+        torch.tensor
+            precision score
+        """
+        # 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
+        )
+
+        if USE_MEAN:
+            tp = torch.sum(true_oh * pred_oh, 0)
+            fp = torch.sum(~true_oh.bool() * pred_oh, 0)
+
+        else:
+            tp = torch.sum(true_oh * pred_oh)
+            fp = torch.sum(~true_oh[pred_oh.bool()].bool())
+
+        return torch.nanmean(tp / (tp + fp))
+
+
+def test_precision_case1():
+    true_precision = 25.0 / 36 if USE_MEAN else 7.0 / 10
+
+    true1 = torch.tensor([0, 1, 2, 1, 0, 2, 1, 0, 2, 1])
+    pred1 = torch.tensor([0, 2, 1, 1, 0, 2, 0, 0, 2, 1])
+    P = Precision(3)
+    precision1 = P(true1, pred1)
+    assert precision1.allclose(torch.tensor(true_precision), atol=1e-5), (
+        f"Precision Score: {precision1.item()}"
+    )
+
+
+def test_precision_case2():
+    true_precision = 8.0 / 15 if USE_MEAN else 6.0 / 15
+
+    true2 = torch.tensor([0, 1, 2, 3, 4, 0, 1, 2, 3, 4, 0, 1, 2, 3, 4])
+    pred2 = torch.tensor([0, 0, 4, 3, 4, 0, 4, 4, 2, 3, 4, 1, 2, 4, 0])
+    P = Precision(5)
+    precision2 = P(true2, pred2)
+    assert precision2.allclose(torch.tensor(true_precision), atol=1e-5), (
+        f"Precision Score: {precision2.item()}"
+    )
+
+
+def test_precision_case3():
+    true_precision = 3.0 / 4 if USE_MEAN else 4.0 / 5
+
+    true3 = torch.tensor([0, 0, 0, 1, 0])
+    pred3 = torch.tensor([1, 0, 0, 1, 0])
+    P = Precision(2)
+    precision3 = P(true3, pred3)
+    assert precision3.allclose(torch.tensor(true_precision), atol=1e-5), (
+        f"Precision Score: {precision3.item()}"
+    )
+
+
+def test_for_zero_denominator():
+    true_precision = 0.0
+    true4 = torch.tensor([1, 1, 1, 1, 1])
+    pred4 = torch.tensor([0, 0, 0, 0, 0])
+    P = Precision(2)
+    precision4 = P(true4, pred4)
+    assert precision4.allclose(torch.tensor(true_precision), atol=1e-5), (
+        f"Precision Score: {precision4.item()}"
+    )
+
+
+if __name__ == "__main__":
+    pass

utils/models/johan_model.py

@@ -0,0 +1,62 @@
+import torch.nn as nn
+
+"""
+Multi-layer perceptron model for image classification.
+"""
+
+# class NeuronLayer(nn.Module):
+#     def __init__(self, in_features, out_features):
+#         super().__init__()
+
+#         self.fc = nn.Linear(in_features, out_features)
+#         self.relu = nn.ReLU()
+
+#     def forward(self, x):
+#         x = self.fc(x)
+#         x = self.relu(x)
+#         return x
+
+
+class JohanModel(nn.Module):
+    """Small MLP model for image classification.
+
+    Parameters
+    ----------
+    in_features : int
+        Numer of input features.
+    num_classes : int
+        Number of classes in the dataset.
+
+    """
+
+    def __init__(self, image_shape, num_classes):
+        super().__init__()
+
+        # Extract features from image shape
+        self.in_channels = image_shape[0]
+        self.height = image_shape[1]
+        self.width = image_shape[2]
+        self.num_classes = num_classes
+        self.in_features = self.in_channels * self.height * self.width
+
+        self.fc1 = nn.Linear(self.in_features, 77)
+        self.fc2 = nn.Linear(77, 77)
+        self.fc3 = nn.Linear(77, 77)
+        self.fc4 = nn.Linear(77, num_classes)
+        self.softmax = nn.Softmax(dim=1)
+        self.relu = nn.ReLU()
+
+    def forward(self, x):
+        for layer in [self.fc1, self.fc2, self.fc3, self.fc4]:
+            x = layer(x)
+            x = self.relu(x)
+        x = self.softmax(x)
+        return x
+
+
+# TODO
+# Add your tests here
+
+
+if __name__ == "__main__":
+    pass  # Add your tests here