Remove unused utilities

Signed-off-by: Edwin Yu <[email protected]>
MemMachine · edwinyyyu · Apr 22, 2026 · Mar 6, 2026 · Mar 9, 2026 · Mar 9, 2026
commit f229f08f39efda762017aae46a976324e6d29668
diff --git a/packages/server/src/memmachine_server/common/utils.py b/packages/server/src/memmachine_server/common/utils.py
@@ -9,11 +9,8 @@
 from datetime import UTC, datetime
 from typing import Any, ParamSpec, TypeVar, cast
 
-import numpy as np
 from nltk import sent_tokenize
 
-from .data_types import SimilarityMetric
-
 T = TypeVar("T")
 P = ParamSpec("P")
 
@@ -272,95 +269,3 @@ def cluster_texts(
         clusters.append(current_cluster)
 
     return clusters
-
-
-def compute_similarity(
-    query_embedding: list[float],
-    candidate_embeddings: list[list[float]],
-    similarity_metric: SimilarityMetric | None = None,
-) -> list[float]:
-    """
-    Compute similarity scores between a query embedding and candidate embeddings.
-
-    Args:
-        query_embedding (list[float]): The embedding of the query.
-        candidate_embeddings (list[list[float]]): A list of candidate embeddings to compare against.
-        similarity_metric (SimilarityMetric | None): The similarity metric to use (default: None).
-
-    Returns:
-        list[float]: A list of similarity scores for each candidate embedding.
-
-    """
-    if not candidate_embeddings:
-        return []
-
-    query_embedding_np = np.array(query_embedding)
-    candidate_embeddings_np = np.array(candidate_embeddings)
-
-    match similarity_metric:
-        case SimilarityMetric.COSINE:
-            magnitude_products = np.linalg.norm(
-                candidate_embeddings_np,
-                axis=-1,
-            ) * np.linalg.norm(query_embedding_np)
-            magnitude_products[magnitude_products == 0] = float("inf")
-
-            scores = (
-                np.dot(candidate_embeddings_np, query_embedding_np) / magnitude_products
-            )
-        case SimilarityMetric.DOT:
-            scores = np.dot(candidate_embeddings_np, query_embedding_np)
-        case SimilarityMetric.EUCLIDEAN:
-            scores = np.linalg.norm(
-                candidate_embeddings_np - query_embedding_np,
-                axis=-1,
-            )
-        case SimilarityMetric.MANHATTAN:
-            scores = np.sum(
-                np.abs(candidate_embeddings_np - query_embedding_np),
-                axis=-1,
-            )
-        case _:
-            # Default to cosine similarity.
-            magnitude_products = np.linalg.norm(
-                candidate_embeddings_np,
-                axis=-1,
-            ) * np.linalg.norm(query_embedding_np)
-            magnitude_products[magnitude_products == 0] = float("inf")
-
-            scores = (
-                np.dot(candidate_embeddings_np, query_embedding_np) / magnitude_products
-            )
-
-    return scores.astype(float).tolist()
-
-
-def similarity_gt(
-    similarity_metric: SimilarityMetric | None = None,
-) -> Callable[[float, float], bool]:
-    """
-    Return a comparator for similarity scores.
-
-    The returned function `gt(a, b)` returns True when score `a`
-    represents greater similarity than score `b` under the given metric.
-
-    For COSINE and DOT, higher scores mean greater similarity (`a > b`).
-    For EUCLIDEAN and MANHATTAN, lower scores mean greater similarity (`a < b`).
-
-    Args:
-        similarity_metric (SimilarityMetric | None):
-            The similarity metric.
-            If None, defaults to greater-than comparison (`a > b`)
-            (default: None).
-
-    Returns:
-        Callable[[float, float], bool]: A comparator for similarity scores.
-
-    """
-    import operator
-
-    match similarity_metric:
-        case SimilarityMetric.EUCLIDEAN | SimilarityMetric.MANHATTAN:
-            return operator.lt
-        case _:
-            return operator.gt