Merge pull request #286 from rudrakatkar/add-tests-metrics

Added  tests for Metrics Module

Author: dpascualhe

tests/test_metrics.py

@@ -0,0 +1,116 @@
+import numpy as np
+import pytest
+from detectionmetrics.utils.metrics import MetricsFactory
+
+
+@pytest.fixture
+def metrics_factory():
+    """Fixture to create a MetricsFactory instance for testing"""
+    return MetricsFactory(n_classes=3)
+
+
+def test_update_confusion_matrix(metrics_factory):
+    """Test confusion matrix updates correctly"""
+    pred = np.array([0, 1, 2, 2, 1])
+    gt = np.array([0, 1, 1, 2, 2])
+    
+    metrics_factory.update(pred, gt)
+    confusion_matrix = metrics_factory.get_confusion_matrix()
+
+    expected = np.array([
+        [1, 0, 0],  # True class 0
+        [0, 1, 1],  # True class 1
+        [0, 1, 1],  # True class 2
+    ])
+    assert np.array_equal(confusion_matrix, expected), "Confusion matrix mismatch"
+
+
+def test_get_tp_fp_fn_tn(metrics_factory):
+    pred = np.array([0, 1, 1, 2, 2])
+    gt = np.array([0, 1, 1, 2, 2])
+    metrics_factory.update(pred, gt)
+
+    assert np.array_equal(metrics_factory.get_tp(), np.array([1, 2, 2]))
+    assert np.array_equal(metrics_factory.get_fp(), np.array([0, 0, 0]))
+    assert np.array_equal(metrics_factory.get_fn(), np.array([0, 0, 0]))
+    assert np.array_equal(metrics_factory.get_tn(), np.array([4, 3, 3]))
+
+def test_recall(metrics_factory):
+    """Test recall calculation"""
+    pred = np.array([0, 1, 2, 2, 1])
+    gt = np.array([0, 1, 1, 2, 2])
+    
+    metrics_factory.update(pred, gt)
+
+    expected_recall = np.array([1.0, 0.5, 0.5])
+    computed_recall = metrics_factory.get_recall()
+    
+    assert np.allclose(computed_recall, expected_recall, equal_nan=True)
+
+def test_accuracy(metrics_factory):
+    """Test global accuracy calculation (non per-class)"""
+    pred = np.array([0, 1, 2, 2, 1])
+    gt = np.array([0, 1, 1, 2, 2])
+
+    metrics_factory.update(pred, gt)
+
+    TP = metrics_factory.get_tp(per_class=False)
+    FP = metrics_factory.get_fp(per_class=False)
+    FN = metrics_factory.get_fn(per_class=False)
+    TN = metrics_factory.get_tn(per_class=False)
+
+    total = TP + FP + FN + TN
+    expected_accuracy = (TP + TN) / total if total > 0 else math.nan
+
+    computed_accuracy = metrics_factory.get_accuracy(per_class=False)
+    assert np.isclose(computed_accuracy, expected_accuracy, equal_nan=True)
+
+def test_f1_score(metrics_factory):
+    """Test F1-score calculation"""
+    pred = np.array([0, 1, 2, 2, 1])
+    gt = np.array([0, 1, 1, 2, 2])
+    
+    metrics_factory.update(pred, gt)
+
+    precision = np.array([1.0, 0.5, 0.5])
+    recall = np.array([1.0, 0.5, 0.5])
+    expected_f1 = 2 * (precision * recall) / (precision + recall)
+
+    computed_f1 = metrics_factory.get_f1_score()
+
+    assert np.allclose(computed_f1, expected_f1, equal_nan=True)
+
+
+def test_edge_cases(metrics_factory):
+    """Test edge cases like empty arrays and division by zero"""
+    pred = np.array([])
+    gt = np.array([])
+
+    with pytest.raises(AssertionError):
+        metrics_factory.update(pred, gt)
+
+    empty_metrics_factory = MetricsFactory(n_classes=3)
+    
+    assert np.isnan(empty_metrics_factory.get_precision(per_class=False))
+    assert np.isnan(empty_metrics_factory.get_recall(per_class=False))
+    assert np.isnan(empty_metrics_factory.get_f1_score(per_class=False))
+    assert np.isnan(empty_metrics_factory.get_iou(per_class=False))
+
+
+def test_macro_micro_weighted(metrics_factory):
+    """Test macro, micro, and weighted metric averaging"""
+    pred = np.array([0, 1, 2, 2, 1])
+    gt = np.array([0, 1, 1, 2, 2])
+    
+    metrics_factory.update(pred, gt)
+
+    macro_f1 = metrics_factory.get_averaged_metric("f1_score", method="macro")
+    micro_f1 = metrics_factory.get_averaged_metric("f1_score", method="micro")
+
+    weights = np.array([0.2, 0.5, 0.3])
+    weighted_f1 = metrics_factory.get_averaged_metric("f1_score", method="weighted", weights=weights)
+
+    assert 0 <= macro_f1 <= 1
+    assert 0 <= micro_f1 <= 1
+    assert 0 <= weighted_f1 <= 1
+