feat: Allow arbitrary values as categoricals/constants/ordinals (#376)

* feat: Allow arbitrary values as categoricals

* fix: Allow constants to have arbitrary values

* feat(Constant): Fix transformation issues with `to_vector`

Author: eddiebergman

changelog.md

@@ -1,3 +1,9 @@
+# Version 1.1.0
+
+* FEAT #376: Allow arbitrary values for `Categorical`, `Ordinal`, and `Constant` hyperparameters.
+* FIX #375: Use `object` dtype for `Constant` np.array of values to prevent numpy type conversions of values.
+
+
 # Version 1.0.1
 
 * FIX #373: Fix `ForbiddenEqualsRelation` when evaluating on vectors of values.
@@ -92,7 +98,7 @@
 
 # Version 0.4.15
 
-* Add `pyproject.toml` to support wheel installation as required in 
+* Add `pyproject.toml` to support wheel installation as required in
   [PEP518](https://medium.com/@grassfedcode/pep-517-and-518-in-plain-english-47208ca8b7a6)
 
 # Version 0.4.14
@@ -120,7 +126,7 @@
 * ADD #135: Add weights to the sampling of categorical hyperparameters.
 * MAINT #129: Performance improvements for the generation of neighbor configurations.
 * MAINT #130: Test the installability of a distribution on travis-ci.
-* FIX #140: Fixes a bug which led to samples lower than the lower bound of 
+* FIX #140: Fixes a bug which led to samples lower than the lower bound of
   `UniformFloatHyperparemeter` if the lower bound was larger than zero and quantization was used.
 * FIX # 138: Fixes a bug in which the readme wasn't read correctly on systems not using UTF8 as
   their default encoding.
@@ -145,29 +151,29 @@
 
 # Version 0.4.9
 
-* Fixes an issue where adding a new forbidden for an unknown hyperparameter 
+* Fixes an issue where adding a new forbidden for an unknown hyperparameter
   did not result in an immediate exception.
 * Add a new argument `vector` to `util.deactivate_inactive_hyperparameters`
 * Make the number of categories a public variable for categorical and
   ordinal hyperparameters
 
 # Version 0.4.8
 
-* Fixes an issue which made serialization of `ForbiddenInCondition` to json 
+* Fixes an issue which made serialization of `ForbiddenInCondition` to json
   fail.
-* MAINT #101: Improved error message on setting illegal value in a 
+* MAINT #101: Improved error message on setting illegal value in a
   configuration.
 * DOC #91: Added a documentation to automl.github.io/ConfigSpace
 
 # Version 0.4.7
 
 * Tests Python3.7.
-* Fixes #87: better handling of Conjunctions when adding them to the 
+* Fixes #87: better handling of Conjunctions when adding them to the
   configuration space.
 * MAINT: Improved type annotation in `util.py` which results in improved
   performance (due to better Cython optimization).
-* MAINT: `util.get_one_exchange_neighborhood` now accepts two arguments 
-  `num_neighbors` and `stdev` which govern the neighborhood creation behaviour 
+* MAINT: `util.get_one_exchange_neighborhood` now accepts two arguments
+  `num_neighbors` and `stdev` which govern the neighborhood creation behaviour
   of several continuous hyperparameters.
 * NEW #85: Add function to obtain active hyperparameters
 * NEW #84: Add field for meta-data to the configuration space object.
@@ -291,21 +297,21 @@
 
 # Version 0.2.1
 
-* FIX: bug which changed order of hyperparameters when adding new 
-  hyperparameter. This was non-deterministic due to the use of dict instead 
+* FIX: bug which changed order of hyperparameters when adding new
+  hyperparameter. This was non-deterministic due to the use of dict instead
   of OrderedDict.
 * FIX: compare configurations with == instead of numpy.allclose.
 * FIX: issue 2, syntax error no longer present during installation
 * FIX: json serialization of configurations and their hyperparameters can now
-       be deserialized by json and still compare equal 
+       be deserialized by json and still compare equal
 
 # Version 0.2
 
-* FIX: bug which made integer values have different float values in the 
+* FIX: bug which made integer values have different float values in the
   underlying vector representation.
-* FIX: bug which could make two configuration spaces compare unequal due to 
+* FIX: bug which could make two configuration spaces compare unequal due to
   the use of defaultdict
-* FEATURE: new feature add_configuration_space, which allows to add a whole 
+* FEATURE: new feature add_configuration_space, which allows to add a whole
   configuration space into an existing configuration space
 * FEATURE: python3.5 support
 * FIX: add function get_parent() to Conjunctions (issue #1)

docs/index.md

@@ -26,6 +26,28 @@ Those are introduced in the [user guide](./guide.md)
     * You can now use your editor to jump to definition and see the source code.
     * Contribute more easily!
 
+    There is no also better support in Categorical, Ordinal and Constant hyperparameters,
+    for arbitrary values, for example:
+
+    ```python
+    from dataclasses import dataclass
+    from ConfigSpace import ConfigurationSpace, Constant
+
+    @dataclass
+    class A:
+        a: int
+
+    def f() -> None:
+        return None
+
+    cs = ConfigurationSpace({
+        "cat": [True, False, None],
+        "othercat": [A(1), f],
+        "constant": Constant("constant": (24, 25)),
+    })
+    ```
+
+
     With this, we have also deprecated many of the previous functions, simplifying the API
     where possible or improving it's clarity. We have tried hard to keep everything backwards
     compatible, and also recommend the new functionality to use!

src/ConfigSpace/api/types/categorical.py

@@ -5,6 +5,7 @@
 from typing_extensions import TypeAlias
 
 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]
@@ -16,7 +17,7 @@ def Categorical(
     name: str,
     items: Sequence[T],
     *,
-    default: T | None = None,
+    default: T | _NotSet = NotSet,
     weights: Sequence[float] | None = None,
     ordered: Literal[False],
     meta: dict | None = None,
@@ -29,7 +30,7 @@ def Categorical(
     name: str,
     items: Sequence[T],
     *,
-    default: T | None = None,
+    default: T | _NotSet = NotSet,
     weights: Sequence[float] | None = None,
     ordered: Literal[True],
     meta: dict | None = None,
@@ -42,7 +43,7 @@ def Categorical(
     name: str,
     items: Sequence[T],
     *,
-    default: T | None = None,
+    default: T | _NotSet = NotSet,
     weights: Sequence[float] | None = None,
     ordered: bool = ...,
     meta: dict | None = None,
@@ -53,7 +54,7 @@ def Categorical(
     name: str,
     items: Sequence[T],
     *,
-    default: T | None = None,
+    default: T | _NotSet = NotSet,
     weights: Sequence[float] | None = None,
     ordered: bool = False,
     meta: dict | None = None,

src/ConfigSpace/conditions.py

@@ -37,21 +37,13 @@
 import numpy as np
 from more_itertools import all_equal, unique_everseen
 
-from ConfigSpace.types import f64
+from ConfigSpace.types import NotSet, f64
 
 if TYPE_CHECKING:
     from ConfigSpace.hyperparameters.hyperparameter import Hyperparameter
     from ConfigSpace.types import Array, Mask
 
 
-class _NotSet:
-    def __repr__(self) -> str:
-        return "ValueNotSetObject"
-
-
-NotSet = _NotSet()  # Sentinal value for unset values
-
-
 class Condition(ABC):
     def __init__(
         self,

src/ConfigSpace/configuration.py

@@ -6,10 +6,9 @@
 
 import numpy as np
 
-from ConfigSpace.conditions import NotSet
 from ConfigSpace.exceptions import IllegalValueError
 from ConfigSpace.hyperparameters import FloatHyperparameter
-from ConfigSpace.types import f64
+from ConfigSpace.types import NotSet, f64
 
 if TYPE_CHECKING:
     from ConfigSpace.configuration_space import ConfigurationSpace

src/ConfigSpace/configuration_space.py

@@ -110,13 +110,9 @@ def _parse_hyperparameters_from_dict(
                 raise ValueError(f"Can't have empty list for categorical {name}")
 
             yield CategoricalHyperparameter(name, hp)
-
-        # If it's an allowed type, it's a constant
-        elif isinstance(hp, (int, str, float)):
-            yield Constant(name, hp)
-
         else:
-            raise ValueError(f"Unknown value '{hp}' for '{name}'")
+            # It's a constant
+            yield Constant(name, hp)
 
 
 class ConfigurationSpace(Mapping[str, Hyperparameter]):