fix: Refrains from using np.ndarray for nested objects (#382)

src/ConfigSpace/api/types/categorical.py

@@ -1,14 +1,13 @@
 from __future__ import annotations
 
 from collections.abc import Sequence
-from typing import Literal, Union, overload
-from typing_extensions import TypeAlias
+from typing import Literal, TypeVar, overload
 
 from ConfigSpace.hyperparameters import CategoricalHyperparameter, OrdinalHyperparameter
 from ConfigSpace.types import NotSet, _NotSet
 
 # We only accept these types in `items`
-T: TypeAlias = Union[str, int, float]
+T = TypeVar("T")
 
 
 # ordered False -> CategoricalHyperparameter

src/ConfigSpace/hyperparameters/categorical.py

@@ -252,10 +252,13 @@ def __init__(
             # This can fail with a ValueError if the choices contain arbitrary objects
             # that are list like.
             seq_choices = np.asarray(choices)
+            if seq_choices.ndim != 1:
+                raise ValueError
 
             # NOTE: Unfortunatly, numpy will promote number types to str
             # if there are string types in the array, where we'd rather
             # stick to object type in that case. Hence the manual...
+            print(seq_choices)
             if seq_choices.dtype.kind in {"U", "S"} and not all(
                 isinstance(choice, str) for choice in choices
             ):

src/ConfigSpace/hyperparameters/hp_components.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from collections.abc import Sequence
 from dataclasses import dataclass, field
 from typing import TYPE_CHECKING, Any, Protocol, TypeVar
 from typing_extensions import override
@@ -202,6 +203,8 @@ def to_vector(self, value: Array[Any]) -> Array[f64]:
 
     @override
     def legal_value(self, value: Array[Any]) -> Mask:
+        print(value)
+        print(value, self.seq)
         if self._lookup is not None:
             return np.array([v in self._lookup for v in value], dtype=np.bool_)
 
@@ -513,6 +516,12 @@ def to_vector(self, value: ObjectArray) -> Array[f64]:
 
     @override
     def to_value(self, vector: Array[f64]) -> ObjectArray:
+        if isinstance(self.value, Sequence) and not isinstance(self.value, str):
+            # We have to convert it into a numpy array of objects carefully
+            # https://stackoverflow.com/a/47389566/5332072
+            _v = np.empty(len(vector), dtype=object)
+            _v[:] = [self.value] * len(vector)
+            return _v
         return np.full_like(vector, self.value, dtype=object)
 
     @override

src/ConfigSpace/hyperparameters/hyperparameter.py

@@ -126,7 +126,7 @@ def __init__(
         self._vector_dist = vector_dist
         self._transformer = transformer
         self._neighborhood = neighborhood
-        self._neighborhood_size = neighborhood_size
+        self._neighborhood_size = neighborhood_size  # type: ignore
         self._value_cast = value_cast
 
         if not self.legal_value(self.default_value):

src/ConfigSpace/hyperparameters/ordinal.py

@@ -94,6 +94,8 @@ def __init__(
             # This can fail with a ValueError if the choices contain arbitrary objects
             # that are list like.
             seq_choices = np.asarray(sequence)
+            if seq_choices.ndim != 1:
+                raise ValueError
 
             # NOTE: Unfortunatly, numpy will promote number types to str
             # if there are string types in the array, where we'd rather
@@ -228,7 +230,7 @@ def pdf_values(self, values: Sequence[Any] | Array[Any]) -> Array[f64]:
             if values.ndim != 1:
                 raise ValueError("Method pdf expects a one-dimensional numpy array")
 
-            vector = self.to_vector(values)
+            vector = self.to_vector(values)  # type: ignore
             return self.pdf_vector(vector)
 
         if self._contains_sequence_as_value:

test/test_hyperparameters.py

@@ -99,6 +99,7 @@ def test_constant():
     # Test getting the size
     for constant in (c1, c2, c3, c4, c5, c1_meta, c6, c7, c8):
         assert constant.size == 1
+        _ = str(constant)  # Ensure str repr works
 
     with pytest.raises(ValueError):
         _ = Constant("value", np.array([1, 2]))
@@ -268,6 +269,7 @@ def test_uniformfloat():
     # Test get_size
     for float_hp in (f1, f3, f4, f5):
         assert np.isinf(float_hp.size)
+        _ = str(float_hp)  # Ensure str repr works
 
 
 def test_uniformfloat_to_integer():
@@ -575,6 +577,7 @@ def test_normalfloat():
     # Test get_size
     for float_hp in (f1, f2, f6):
         assert np.isinf(float_hp.size)
+        _ = str(float_hp)  # Ensure str repr works
 
     with pytest.raises(ValueError):
         _ = NormalFloatHyperparameter(
@@ -2147,6 +2150,11 @@ def test_categorical():
     assert f5.size == 1000
     assert f6.size == 2
 
+    fn = CategoricalHyperparameter("param", [("a", "b"), [0, 1]])
+    assert fn.default_value == ("a", "b")
+    assert fn.legal_value(("a", "b"))
+    _ = str(fn)
+
 
 def test_cat_equal():
     # Test that weights are properly normalized and compared
@@ -2756,6 +2764,14 @@ def test_ordinal_is_legal():
     assert not f1.legal_vector("Hahaha")  # type: ignore
 
 
+def test_ordinal_nested_lists_prints_correctly():
+    f1 = OrdinalHyperparameter(
+        "temp",
+        [["freezing", "cold", "warm", "hot"], ["a", "b"]],
+    )
+    _ = str(f1)
+
+
 def test_ordinal_check_order():
     f1 = OrdinalHyperparameter("temp", ["freezing", "cold", "warm", "hot"])
     assert f1.check_order("freezing", "cold")