Add roc_auc_score + tests

src/metrics/classification.py

@@ -94,7 +94,6 @@ def f1_score(y_true: np.ndarray, y_pred: np.ndarray) -> float:
 def roc_curve(y_true: np.ndarray, y_pred: np.ndarray) -> (np.ndarray, np.ndarray):
     """
     Calculate ROC curve values.
-
     :param y_true: Target labels (n_samples, ).
     :param y_pred: Predictions probability (n_samples, ).
     :return:
@@ -130,3 +129,21 @@ def roc_curve(y_true: np.ndarray, y_pred: np.ndarray) -> (np.ndarray, np.ndarray
             fpr[i] = fp / (tn + fp)
 
     return tpr, fpr, thresholds
+
+
+def roc_auc_score(y_true: np.ndarray, y_pred: np.ndarray) -> float:
+    """
+    Calculate ROC-AUC score using trapezoidal rule.
+    :param y_true: Target labels (n_samples, ).
+    :param y_pred: Predictions probability (n_samples, ).
+    :return: ROC-AUC score.
+    """
+
+    tpr, fpr, _ = roc_curve(y_true, y_pred)
+
+    shifted_tpr = np.roll(tpr, shift=1)
+    shifted_tpr[0] = 0
+
+    height = np.diff(fpr, n=1, prepend=0)
+    auc = np.sum((tpr + shifted_tpr) / 2 * height)
+    return auc.item()

tests/metrics/classifciation/test_roc_auc_score.py

@@ -0,0 +1,53 @@
+import numpy as np
+
+from src.metrics import roc_auc_score
+
+
+def test_roc_auc_identical_labels():
+    y_true = np.array([1, 0, 1, 0, 1])
+    y_pred = np.array([1, 0, 1, 0, 1])
+
+    expected_roc_auc = 1.0
+    roc_auc = roc_auc_score(y_true, y_pred)
+
+    assert np.isclose(expected_roc_auc, roc_auc, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_auc_reversed_labels():
+    y_true = np.array([1, 0, 1, 0, 1])
+    y_pred = np.array([0, 1, 0, 1, 0])
+
+    expected_roc_auc = 0.0
+    roc_auc = roc_auc_score(y_true, y_pred)
+
+    assert np.isclose(expected_roc_auc, roc_auc, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_auc_all_true():
+    y_true = np.array([1, 0, 0, 0, 1])
+    y_pred = np.array([1, 1, 1, 1, 1])
+
+    expected_roc_auc = 0.5
+    roc_auc = roc_auc_score(y_true, y_pred)
+
+    np.isclose(expected_roc_auc, roc_auc, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_auc_equal_prob():
+    y_true = np.array([1, 0, 0, 0, 1])
+    y_pred = np.array([0.5, 0.5, 0.5, 0.5, 0.5])
+
+    expected_roc_auc = 0.5
+    roc_auc = roc_auc_score(y_true, y_pred)
+
+    np.isclose(expected_roc_auc, roc_auc, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_auc_close_to_target():
+    y_true = np.array([1, 0, 0, 0, 1])
+    y_pred = np.array([0.9, 0.1, 0.1, 0.1, 0.9])
+
+    expected_roc_auc = 1.0
+    roc_auc = roc_auc_score(y_true, y_pred)
+
+    np.isclose(expected_roc_auc, roc_auc, atol=1e-5, rtol=1e-5)