scikit-learn · adrinjalali · Apr 16, 2024 · Oct 7, 2020 · Jan 16, 2021 · Jan 19, 2021
diff --git a/doc/whats_new/v1.5.rst b/doc/whats_new/v1.5.rst
@@ -339,6 +339,10 @@ Changelog
   in favor of `y_proba`. `y_prob` will be removed in version 1.7.
   :pr:`28092` by :user:`Adam Li <adam2392>`.
 
+- |API| For classifiers and classification metrics, labels encoded as bytes
+  is deprecated and will raise an error in v1.6.
+  :pr:`18555` by :user:`Kaushik Amar Das <cozek>`.
+
 :mod:`sklearn.mixture`
 ......................
 
@@ -418,6 +422,11 @@ Changelog
 - |API| :func:`utils.tosequence` is deprecated and will be removed in version 1.7.
   :pr:`28763` by :user:`Jérémie du Boisberranger <jeremiedbb>`.
 
+- |API| Raise informative warning message in :func:`type_of_target` when
+  represented as bytes. For classifiers and classification metrics, labels encoded
+  as bytes is deprecated and will raise an error in v1.6.
+  :pr:`18555` by :user:`Kaushik Amar Das <cozek>`.
+
 - |Fix| :func:`~utils._safe_indexing` now works correctly for polars DataFrame when
   `axis=0` and supports indexing polars Series.
   :pr:`28521` by :user:`Yao Xiao <Charlie-XIAO>`.

diff --git a/sklearn/metrics/tests/test_ranking.py b/sklearn/metrics/tests/test_ranking.py
@@ -29,10 +29,12 @@
 from sklearn.preprocessing import label_binarize
 from sklearn.random_projection import _sparse_random_matrix
 from sklearn.utils._testing import (
+    _convert_container,
     assert_allclose,
     assert_almost_equal,
     assert_array_almost_equal,
     assert_array_equal,
+    ignore_warnings,
 )
 from sklearn.utils.extmath import softmax
 from sklearn.utils.fixes import CSR_CONTAINERS
@@ -864,17 +866,6 @@ def test_binary_clf_curve_implicit_pos_label(curve_func):
     with pytest.raises(ValueError, match=msg):
         curve_func(np.array(["a", "b"], dtype=object), [0.0, 1.0])
 
-    # The error message is slightly different for bytes-encoded
-    # class labels, but otherwise the behavior is the same:
-    msg = (
-        "y_true takes value in {b'a', b'b'} and pos_label is "
-        "not specified: either make y_true take "
-        "value in {0, 1} or {-1, 1} or pass pos_label "
-        "explicitly."
-    )
-    with pytest.raises(ValueError, match=msg):
-        curve_func(np.array([b"a", b"b"], dtype="<S1"), [0.0, 1.0])
-
     # Check that it is possible to use floating point class labels
     # that are interpreted similarly to integer class labels:
     y_pred = [0.0, 1.0, 0.2, 0.42]
@@ -884,6 +875,23 @@ def test_binary_clf_curve_implicit_pos_label(curve_func):
         np.testing.assert_allclose(int_curve_part, float_curve_part)
 
 
+# TODO(1.5): Update test to check for error when bytes support is removed.
+@ignore_warnings(category=FutureWarning)
+@pytest.mark.parametrize("curve_func", [precision_recall_curve, roc_curve])
+@pytest.mark.parametrize("labels_type", ["list", "array"])
+def test_binary_clf_curve_implicit_bytes_pos_label(curve_func, labels_type):
+    # Check that using bytes class labels raises an informative
+    # error for any supported string dtype:
+    labels = _convert_container([b"a", b"b"], labels_type)
+    msg = (
+        "y_true takes value in {b'a', b'b'} and pos_label is not "
+        "specified: either make y_true take value in {0, 1} or "
+        "{-1, 1} or pass pos_label explicitly."
+    )
+    with pytest.raises(ValueError, match=msg):
+        curve_func(labels, [0.0, 1.0])
+
+
 @pytest.mark.parametrize("curve_func", CURVE_FUNCS)
 def test_binary_clf_curve_zero_sample_weight(curve_func):
     y_true = [0, 0, 1, 1, 1]

diff --git a/sklearn/utils/multiclass.py b/sklearn/utils/multiclass.py
@@ -342,13 +342,24 @@ def type_of_target(y, input_name=""):
                 # see NEP 34
                 y = check_array(y, dtype=object, **check_y_kwargs)
 
-    # The old sequence of sequences format
     try:
-        first_row = y[[0], :] if issparse(y) else y[0]
+        # TODO(1.7): Change to ValueError when byte labels is deprecated.
+        # labels in bytes format
+        first_row_or_val = y[[0], :] if issparse(y) else y[0]
+        if isinstance(first_row_or_val, bytes):
+            warnings.warn(
+                (
+                    "Support for labels represented as bytes is deprecated in v1.5 and"
+                    " will error in v1.7. Convert the labels to a string or integer"
+                    " format."
+                ),
+                FutureWarning,
+            )
+        # The old sequence of sequences format
         if (
-            not hasattr(first_row, "__array__")
-            and isinstance(first_row, Sequence)
-            and not isinstance(first_row, str)
+            not hasattr(first_row_or_val, "__array__")
+            and isinstance(first_row_or_val, Sequence)
+            and not isinstance(first_row_or_val, str)
         ):
             raise ValueError(
                 "You appear to be using a legacy multi-label data"
@@ -390,9 +401,9 @@ def type_of_target(y, input_name=""):
             return "continuous" + suffix
 
     # Check multiclass
-    if issparse(first_row):
-        first_row = first_row.data
-    if xp.unique_values(y).shape[0] > 2 or (y.ndim == 2 and len(first_row) > 1):
+    if issparse(first_row_or_val):
+        first_row_or_val = first_row_or_val.data
+    if xp.unique_values(y).shape[0] > 2 or (y.ndim == 2 and len(first_row_or_val) > 1):
         # [1, 2, 3] or [[1., 2., 3]] or [[1, 2]]
         return "multiclass" + suffix
     else:

diff --git a/sklearn/utils/tests/test_multiclass.py b/sklearn/utils/tests/test_multiclass.py
@@ -10,6 +10,7 @@
 from sklearn.utils._array_api import yield_namespace_device_dtype_combinations
 from sklearn.utils._testing import (
     _array_api_for_tests,
+    _convert_container,
     assert_allclose,
     assert_array_almost_equal,
     assert_array_equal,
@@ -595,3 +596,18 @@ def test_ovr_decision_function():
     ]
 
     assert_allclose(dec_values, dec_values_one, atol=1e-6)
+
+
+# TODO(1.6): Change to ValueError when byte labels is deprecated.
+@pytest.mark.parametrize("input_type", ["list", "array"])
+def test_labels_in_bytes_format(input_type):
+    # check that we raise an error with bytes encoded labels
+    # non-regression test for:
+    # https://github.com/scikit-learn/scikit-learn/issues/16980
+    target = _convert_container([b"a", b"b"], input_type)
+    err_msg = (
+        "Support for labels represented as bytes is deprecated in v1.5 and will"
+        " error in v1.7. Convert the labels to a string or integer format."
+    )
+    with pytest.warns(FutureWarning, match=err_msg):
+        type_of_target(target)