diff --git a/setup.py b/setup.py
index 9b8897233..d5b82ec68 100755
--- a/setup.py
+++ b/setup.py
@@ -30,6 +30,7 @@ MOD_NAMES = [
     "spacy.lexeme",
     "spacy.vocab",
     "spacy.attrs",
+    "spacy.kb.candidate",
     "spacy.kb.kb",
     "spacy.kb.kb_in_memory",
     "spacy.ml.tb_framework",
diff --git a/spacy/kb/candidate.py b/spacy/kb/candidate.py
deleted file mode 100644
index 9778b2880..000000000
--- a/spacy/kb/candidate.py
+++ /dev/null
@@ -1,116 +0,0 @@
-import abc
-from typing import List, Union, Callable
-
-from ..errors import Errors
-
-
-class Candidate(abc.ABC):
-    """A `Candidate` object refers to a textual mention that may or may not be resolved
-    to a specific `entity_id` from a Knowledge Base. This will be used as input for the entity_id linking
-    algorithm which will disambiguate the various candidates to the correct one.
-    Each candidate (mention, entity_id) pair is assigned a certain prior probability.
-
-    DOCS: https://spacy.io/api/kb/#candidate-init
-    """
-
-    def __init__(
-        self,
-        mention: str,
-        entity_id: Union[str, int],
-        entity_vector: List[float],
-        prior_prob: float,
-    ):
-        """Initializes properties of `Candidate` instance.
-        mention (str): Mention text for this candidate.
-        entity_id (Union[str, int]): Unique entity ID.
-        entity_vector (List[float]): Entity embedding.
-        prior_prob (float): Prior probability of entity for this mention - i.e. the probability that, independent of
-            the context, this mention resolves to this entity_id in the corpus used to build the knowledge base. In
-            cases in which this isn't always possible (e.g.: the corpus to analyse contains mentions that the KB corpus
-            doesn't) it might be better to eschew this information and always supply the same value.
-        """
-        self._mention = mention
-        self._entity_id = entity_id
-        # Note that hashing an int value yields the same int value.
-        self._entity_id_hash = hash(entity_id)
-        self._entity_vector = entity_vector
-        self._prior_prob = prior_prob
-
-    @property
-    def entity_id(self) -> Union[str, int]:
-        """RETURNS (Union[str, int]): Unique entity ID."""
-        return self._entity_id
-
-    @property
-    def entity_id_int(self) -> int:
-        """RETURNS (int): Numerical representation of entity ID (if entity ID is numerical, this is just the entity ID,
-        otherwise the hash of the entity ID string)."""
-        return self._entity_id_hash
-
-    @property
-    def entity_id_str(self) -> str:
-        """RETURNS (str): String representation of entity ID."""
-        return str(self._entity_id)
-
-    @property
-    def mention(self) -> str:
-        """RETURNS (str): Mention."""
-        return self._mention
-
-    @property
-    def entity_vector(self) -> List[float]:
-        """RETURNS (List[float]): Entity vector."""
-        return self._entity_vector
-
-    @property
-    def prior_prob(self) -> float:
-        """RETURNS (List[float]): Entity vector."""
-        return self._prior_prob
-
-
-class InMemoryCandidate(Candidate):
-    """Candidate for InMemoryLookupKB."""
-
-    def __init__(
-        self,
-        hash_to_str: Callable[[int], str],
-        entity_id: int,
-        mention: str,
-        entity_vector: List[float],
-        prior_prob: float,
-        entity_freq: int,
-    ):
-        """
-        hash_to_str (Callable[[int], str]): Callable retrieving entity name from provided entity/vocab hash.
-        entity_id (str): Entity ID as hash that can be looked up with InMemoryKB.vocab.strings.__getitem__().
-        entity_freq (int): Entity frequency in KB corpus.
-        entity_vector (List[float]): Entity embedding.
-        mention (str): Mention.
-        prior_prob (float): Prior probability of entity for this mention - i.e. the probability that, independent of
-            the context, this mention resolves to this entity_id in the corpus used to build the knowledge base. In
-            cases in which this isn't always possible (e.g.: the corpus to analyse contains mentions that the KB corpus
-            doesn't) it might be better to eschew this information and always supply the same value.
-        """
-        super().__init__(
-            mention=mention,
-            entity_id=entity_id,
-            entity_vector=entity_vector,
-            prior_prob=prior_prob,
-        )
-        self._hash_to_str = hash_to_str
-        self._entity_freq = entity_freq
-        if not isinstance(self._entity_id, int):
-            raise ValueError(
-                Errors.E4005.format(should_type="int", is_type=str(type(entity_id)))
-            )
-        self._entity_id_str = self._hash_to_str(self._entity_id)
-
-    @property
-    def entity_freq(self) -> float:
-        """RETURNS (float): Relative entity frequency."""
-        return self._entity_freq
-
-    @property
-    def entity_id_str(self) -> str:
-        """RETURNS (str): String representation of entity ID."""
-        return self._entity_id_str