fix: handle read-only arrays in DunedinPACE quantile normalization (#195)

biolearn/dunedin_pace.py

@@ -50,6 +50,9 @@ def quantile_normalize_using_target(data, target_values):
     """
     Apply quantile normalization on data using target values.
     """
+    # Writable float copy: DataFrame.values can be read-only under pandas>=3.0
+    # copy-on-write, which would break the in-place updates below.
+    data = np.array(data, dtype=float)
     sorted_target = np.sort(target_values)
 
     for _, column_data in enumerate(data.T):

biolearn/test/test_dunedin_pace.py

@@ -0,0 +1,41 @@
+import numpy as np
+import pandas as pd
+
+from biolearn.dunedin_pace import quantile_normalize_using_target
+
+
+def _target():
+    return list(np.linspace(0.0, 1.0, 10))
+
+
+def test_quantile_normalize_accepts_read_only_array():
+    # Arrays from DataFrame.values can be read-only (e.g. pandas
+    # copy-on-write in pandas >= 3.0). Normalization must not fail with
+    # "assignment destination is read-only". See issue #195.
+    data = pd.DataFrame(np.random.rand(10, 3)).values
+    data.flags.writeable = False
+
+    result = quantile_normalize_using_target(data, _target())
+
+    assert result.shape == (10, 3)
+
+
+def test_quantile_normalize_does_not_mutate_input():
+    data = pd.DataFrame(np.random.rand(10, 3)).values
+    original = data.copy()
+
+    quantile_normalize_using_target(data, _target())
+
+    np.testing.assert_array_equal(data, original)
+
+
+def test_quantile_normalize_maps_values_onto_target_distribution():
+    target = _target()
+    data = pd.DataFrame(np.random.rand(10, 3)).values
+
+    result = quantile_normalize_using_target(data, target)
+
+    sorted_target = np.sort(target)
+    midpoints = 0.5 * (sorted_target[:-1] + sorted_target[1:])
+    allowed = np.round(np.concatenate([sorted_target, midpoints]), 6)
+    assert np.all(np.isin(np.round(result, 6), allowed))