adding test for PR_AUC, FAR and FNR

tests/test_metrics.py

@@ -5,13 +5,14 @@
 import numpy as np
 import pytest
 import torch
-from mfai.torch.metrics import CSINeighborood
+from mfai.torch.metrics import CSINeighborood, PR_AUC, FAR, FNR
 
 
-@pytest.mark.parametrize("num_neighbors,expected", [(0, 0.36), (1, 0.91)])
-def test_csi_binary(num_neighbors: int, expected: float):
+@pytest.mark.parametrize("num_neighbors,expected_value", [(0, 0.36), (1, 0.91)])
+def test_csi_binary(num_neighbors: int, expected_value: float):
     """
-    Build two 5x5 tensors and compute the CSI for binary task.
+    Build tensors of size (2, 1, 5, 5), compute the CSI for binary task and check if the output is
+    the result expected.
     """
     y_true = torch.tensor(
         np.array(
@@ -66,15 +67,14 @@ def test_csi_binary(num_neighbors: int, expected: float):
     csi.update(preds=y_hat, targets=y_true)
     csi_score = torch.round(csi.compute(), decimals=2).float()
 
-    assert (
-        csi_score - expected < 0.01
-    ), f"Failed to compute the CSI, return {csi_score} instead of {expected}."
+    assert pytest.approx(csi_score, 0.001) == expected_value
 
 
-@pytest.mark.parametrize("num_neighbors,expected", [(0, 0.43), (1, 0.79)])
-def test_csi_multiclass(num_neighbors: int, expected: torch.Tensor):
+@pytest.mark.parametrize("num_neighbors,expected_value", [(0, 0.43), (1, 0.79)])
+def test_csi_multiclass(num_neighbors: int, expected_value: torch.Tensor):
     """
-    Build two 5x5 tensors and compute the CSI for binary task.
+    Build tensors of size (1, 1, 5, 5), compute the CSI for multiclass taskand check if the output is
+    the result expected.
     """
     y_true = torch.tensor(
         np.array(
@@ -118,15 +118,14 @@ def test_csi_multiclass(num_neighbors: int, expected: torch.Tensor):
     print(csi.false_negatives)
     print(csi_score)
 
-    assert (
-        torch.sum(csi_score - expected) < 0.01
-    ), f"Failed to compute the CSI, return {csi_score} instead of {expected}."
+    assert pytest.approx(csi_score, 0.001) == expected_value
 
 
-@pytest.mark.parametrize("num_neighbors,expected", [(0, 0.36), (1, 0.81)])
-def test_csi_multilabel(num_neighbors: int, expected: torch.Tensor):
+@pytest.mark.parametrize("num_neighbors,expected_value", [(0, 0.36), (1, 0.81)])
+def test_csi_multilabel(num_neighbors: int, expected_value: torch.Tensor):
     """
-    Build two 5x5 tensors and compute the CSI for binary task.
+    Build tensors of size (1, 3, 5, 5), compute the CSI for multilabel task and check if the output is
+    the result expected.
     """
     y_true = torch.tensor(
         np.array(
@@ -191,6 +190,43 @@ def test_csi_multilabel(num_neighbors: int, expected: torch.Tensor):
     csi.update(preds=y_hat, targets=y_true)
     csi_score = torch.round(csi.compute(), decimals=2).float()
 
-    assert (
-        torch.sum(csi_score - expected) < 0.001
-    ), f"Failed to compute the CSI, return {csi_score} instead of {expected}."
+    assert pytest.approx(csi_score, 0.001) == expected_value
+
+
+def test_pr_auc():
+    """
+    Test of the compute of the Precision-Recall Area Under the Curve.
+    """
+    preds = torch.tensor([0.0, 1.0, 0.0, 1.0])
+    targets = torch.tensor([0, 0, 1, 1])
+    far = PR_AUC()
+    far.update(preds, targets)
+    auc_value = far.compute()
+    expected_value = 0.125
+    assert pytest.approx(auc_value, 0.001) == expected_value
+
+
+def test_far():
+    """
+    Test of the compute of the False Alarm Rate.
+    """
+    preds = torch.tensor([0.0, 1.0, 0.0, 1.0])
+    targets = torch.tensor([0, 0, 1, 1])
+    far = FAR("binary")
+    far.update(preds, targets)
+    auc_value = far.compute()
+    expected_value = 0.5
+    assert pytest.approx(auc_value, 0.001) == expected_value
+
+
+def test_fnr():
+    """
+    Test of the compute of the False Alarm Rate.
+    """
+    preds = torch.tensor([0.0, 1.0, 0.0, 1.0])
+    targets = torch.tensor([0, 0, 1, 1])
+    fnr = FNR()
+    fnr.update(preds, targets)
+    auc_value = fnr.compute()
+    expected_value = 0.5
+    assert pytest.approx(auc_value, 0.001) == expected_value