Categorical search space tweaks (#869)

tests/unit/test_space.py

@@ -38,6 +38,7 @@
     SearchSpace,
     TaggedMultiSearchSpace,
     TaggedProductSearchSpace,
+    cast_encoder,
     one_hot_encoder,
 )
 from trieste.types import TensorType
@@ -1759,6 +1760,11 @@ def test_categorical_search_space__to_tags_raises_for_non_integers() -> None:
             tf.constant([[0], [0]], dtype=tf.float64),
             tf.constant([[1], [1]], dtype=tf.float64),
         ),
+        (
+            CategoricalSearchSpace(["Y", "N"]),
+            tf.constant([[0], [1], [0]], dtype=tf.float64),
+            tf.constant([[0], [1], [0]], dtype=tf.float64),
+        ),
         (
             CategoricalSearchSpace(["R", "G", "B"], dtype=tf.float32),
             tf.constant([[0], [2], [1]], dtype=tf.float32),
@@ -1777,13 +1783,13 @@ def test_categorical_search_space__to_tags_raises_for_non_integers() -> None:
         (
             CategoricalSearchSpace([["R", "G", "B"], ["Y", "N"]]),
             tf.constant([[0, 0], [2, 0], [1, 1]], dtype=tf.float64),
-            tf.constant([[1, 0, 0, 1, 0], [0, 0, 1, 1, 0], [0, 1, 0, 0, 1]], dtype=tf.float64),
+            tf.constant([[1, 0, 0, 0], [0, 0, 1, 0], [0, 1, 0, 1]], dtype=tf.float64),
         ),
         (
             CategoricalSearchSpace([["R", "G", "B"], ["Y", "N"]]),
             tf.constant([[[0, 0], [0, 0]], [[2, 0], [1, 1]]], dtype=tf.float64),
             tf.constant(
-                [[[1, 0, 0, 1, 0], [1, 0, 0, 1, 0]], [[0, 0, 1, 1, 0], [0, 1, 0, 0, 1]]],
+                [[[1, 0, 0, 0], [1, 0, 0, 0]], [[0, 0, 1, 0], [0, 1, 0, 1]]],
                 dtype=tf.float64,
             ),
         ),
@@ -1824,6 +1830,12 @@ def test_categorical_search_space_one_hot_encoding(
         pytest.param(
             CategoricalSearchSpace(["Y", "N"]),
             tf.constant([[0], [2], [1]]),
+            ValueError,
+            id="Out of range binary input value",
+        ),
+        pytest.param(
+            CategoricalSearchSpace(["Y", "N", "maybe"]),
+            tf.constant([[0], [3], [1]]),
             InvalidArgumentError,
             id="Out of range input value",
         ),
@@ -1859,3 +1871,19 @@ def test_unbound_search_spaces(
         space.lower
     with pytest.raises(AttributeError):
         space.upper
+
+
+@pytest.mark.parametrize("input_dtype", [None, tf.float64, tf.float32])
+@pytest.mark.parametrize("output_dtype", [None, tf.float64, tf.float32])
+def test_cast_encoder(input_dtype: Optional[tf.DType], output_dtype: Optional[tf.DType]) -> None:
+
+    query_points = tf.constant([1, 2, 3], dtype=tf.int32)
+
+    def add_encoder(x: TensorType) -> TensorType:
+        assert x.dtype is (input_dtype or tf.int32)
+        return x + 1
+
+    encoder = cast_encoder(add_encoder, input_dtype=input_dtype, output_dtype=output_dtype)
+    points = encoder(query_points)
+    assert points.dtype is (output_dtype or input_dtype or tf.int32)
+    npt.assert_array_equal(tf.cast(query_points + 1, points.dtype), points)

trieste/space.py

@@ -518,6 +518,24 @@ def one_hot_encoder(space: SearchSpace) -> EncoderFunction:
     return space.one_hot_encoder if isinstance(space, HasOneHotEncoder) else lambda x: x
 
 
+def cast_encoder(
+    encoder: EncoderFunction,
+    input_dtype: Optional[tf.DType] = None,
+    output_dtype: Optional[tf.DType] = None,
+) -> EncoderFunction:
+    "A utility function for casting the input and/or output of an encoder."
+
+    def cast_and_encode(x: TensorType) -> TensorType:
+        if input_dtype is not None:
+            x = tf.cast(x, input_dtype)
+        y = encoder(x)
+        if output_dtype is not None:
+            y = tf.cast(y, output_dtype)
+        return y
+
+    return cast_and_encode
+
+
 def one_hot_encoded_space(space: SearchSpace) -> SearchSpace:
     "A bounded search space corresponding to the one-hot encoding of the given space."
 
@@ -633,7 +651,14 @@ def tags(self) -> Sequence[Sequence[str]]:
 
     @property
     def one_hot_encoder(self) -> EncoderFunction:
-        """A one-hot encoder for the numerical indices."""
+        """A one-hot encoder for the numerical indices. Note that binary categories
+        are left unchanged instead of adding an unnecessary second feature."""
+
+        def binary_encoder(x: TensorType) -> TensorType:
+            # no need to one-hot encode binary categories (but we should still validate)
+            if tf.reduce_any((x != 0) & (x != 1)):
+                raise ValueError(f"Invalid values {tf.boolean_mask(x, ((x != 0) & (x != 1)))}")
+            return x
 
         def encoder(x: TensorType) -> TensorType:
             flat_x, unflatten = flatten_leading_dims(x)
@@ -644,7 +669,11 @@ def encoder(x: TensorType) -> TensorType:
                 )
             columns = tf.split(flat_x, flat_x.shape[-1], axis=1)
             encoders = [
-                tf_keras.layers.CategoryEncoding(num_tokens=len(ts), output_mode="one_hot")
+                (
+                    binary_encoder
+                    if len(ts) == 2
+                    else tf_keras.layers.CategoryEncoding(num_tokens=len(ts), output_mode="one_hot")
+                )
                 for ts in self.tags
             ]
             encoded = tf.concat(