Add roc_curve + test

src/metrics/classification.py

@@ -89,3 +89,39 @@ def f1_score(y_true: np.ndarray, y_pred: np.ndarray) -> float:
 
     f1 = 2 * precision * recall / (precision + recall)
     return f1
+
+
+def roc_curve(y_true: np.ndarray, y_pred: np.ndarray, threshold_step: float = 0.1) -> (np.ndarray, np.ndarray):
+    """
+    Calculate ROC curve values.
+
+    :param y_true: Target labels (n_samples, ).
+    :param y_pred: Target predictions probabilities (n_samples, ).
+    :param threshold_step: Threshold step for calculating rates (float).
+    :return: True positive rates (threshold_steps, ), False positive rates (threshold_steps, ).
+    """
+    thresholds = np.arange(0.1, 1, threshold_step)
+
+    tpr = np.zeros_like(thresholds)
+    fpr = np.zeros_like(thresholds)
+
+    for i, threshold in enumerate(thresholds):
+        predictions = np.where(y_pred >= threshold, 1, 0)
+
+        # Calculate true positive rates.
+        true_positives = np.sum(y_true * predictions)
+        false_negatives = np.sum(y_true * (1 - predictions))
+        if true_positives == 0 and false_negatives == 0:
+            tpr[i] = 0.0
+        else:
+            tpr[i] = true_positives / (true_positives + false_negatives)
+
+        # Calculate false positive rates.
+        true_negatives = np.sum((1 - y_true) * (1 - predictions))
+        false_positives = np.sum((1 - y_true) * predictions)
+        if false_positives == 0 and true_negatives == 0:
+            fpr[i] = 0.0
+        else:
+            fpr[i] = true_negatives / (true_negatives + false_positives)
+
+    return tpr, fpr

tests/metrics/classifciation/test_roc_curve.py

@@ -0,0 +1,94 @@
+import numpy as np
+
+from src.metrics import roc_curve
+
+
+def test_roc_curve_identical_labels():
+    y_true = np.array([1, 0, 1, 0, 1])
+    y_pred = np.array([1, 0, 1, 0, 1])
+
+    expected_tpr = np.ones(shape=(9,))
+    expected_fpr = np.ones(shape=(9,))
+
+    tpr, fpr = roc_curve(y_true, y_pred)
+
+    assert np.allclose(expected_tpr, tpr, atol=1e-5, rtol=1e-5)
+    assert np.allclose(expected_fpr, fpr, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_curve_different_labels():
+    y_true = np.array([1, 1, 1, 1, 1])
+    y_pred = np.array([0, 0, 0, 0, 0])
+
+    expected_tpr = np.zeros(shape=(9,))
+    expected_fpr = np.zeros(shape=(9,))
+
+    tpr, fpr = roc_curve(y_true, y_pred)
+
+    assert np.allclose(expected_tpr, tpr, atol=1e-5, rtol=1e-5)
+    assert np.allclose(expected_fpr, fpr, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_curve_reversed_different_labels():
+    y_true = np.array([0, 0, 0, 0, 0])
+    y_pred = np.array([1, 1, 1, 1, 1])
+
+    expected_tpr = np.zeros(shape=(9,))
+    expected_fpr = np.zeros(shape=(9,))
+
+    tpr, fpr = roc_curve(y_true, y_pred)
+
+    assert np.allclose(expected_tpr, tpr, atol=1e-5, rtol=1e-5)
+    assert np.allclose(expected_fpr, fpr, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_curve_reversed_labels():
+    y_true = np.array([1, 0, 1, 0, 1])
+    y_pred = np.array([0, 1, 0, 1, 0])
+
+    expected_tpr = np.zeros(shape=(9,))
+    expected_fpr = np.zeros(shape=(9,))
+
+    tpr, fpr = roc_curve(y_true, y_pred)
+
+    assert np.allclose(expected_tpr, tpr, atol=1e-5, rtol=1e-5)
+    assert np.allclose(expected_fpr, fpr, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_cruve_all_true():
+    y_true = np.array([1, 0, 0, 0, 1])
+    y_pred = np.array([1, 1, 1, 1, 1])
+
+    expected_tpr = np.ones(shape=(9,))
+    expected_fpr = np.zeros(shape=(9,))
+
+    tpr, fpr = roc_curve(y_true, y_pred)
+
+    assert np.allclose(expected_tpr, tpr, atol=1e-5, rtol=1e-5)
+    assert np.allclose(expected_fpr, fpr, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_curve_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_tpr = np.array([1, 1, 1, 1, 1, 0, 0, 0, 0])
+    expected_fpr = np.array([0, 0, 0, 0, 0, 1, 1, 1, 1])
+
+    tpr, fpr = roc_curve(y_true, y_pred)
+
+    assert np.allclose(expected_tpr, tpr, atol=1e-5, rtol=1e-5)
+    assert np.allclose(expected_fpr, fpr, atol=1e-5, rtol=1e-5)
+
+
+def test_roc_curve_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_tpr = np.array([1, 1, 1, 1, 1, 1, 1, 1, 1])
+    expected_fpr = np.array([0, 1, 1, 1, 1, 1, 1, 1, 1])
+
+    tpr, fpr = roc_curve(y_true, y_pred)
+
+    assert np.allclose(expected_tpr, tpr, atol=1e-5, rtol=1e-5)
+    assert np.allclose(expected_fpr, fpr, atol=1e-5, rtol=1e-5)