Added compute functions

pysentence_similarity/compute.py

@@ -0,0 +1,236 @@
+"""Compute functions module."""
+import logging
+import numpy as np
+
+# Set up logging
+logger = logging.getLogger("pysentence-similarity:compute_function")
+
+
+def cosine(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray
+) -> float:
+    """Compute cosine similarity between two embedding vectors.
+
+    :param embedding_1: First embedding vector.
+    :type embedding_1: np.ndarray
+    :param embedding_2: Second embedding vector.
+    :type embedding_2: np.ndarray
+    :return: Cosine similarity score.
+    :rtype: float
+    """
+    try:
+        embedding_1_norm = embedding_1 / np.linalg.norm(embedding_1)
+        embedding_2_norm = embedding_2 / np.linalg.norm(embedding_2)
+        similarity = np.dot(embedding_1_norm, embedding_2_norm)
+        return similarity
+    except Exception as err:
+        logger.error("Error computing cosine similarity: %s", err)
+        raise
+
+
+def euclidean(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray
+) -> float:
+    """Compute Euclidean distance between two embedding vectors.
+
+    :param embedding_1: First embedding vector.
+    :type embedding_1: np.ndarray
+    :param embedding_2: Second embedding vector.
+    :type embedding_2: np.ndarray
+    :return: Euclidean distance.
+    :rtype: float
+    """
+    try:
+        distance = np.linalg.norm(embedding_1 - embedding_2)
+        return distance
+    except Exception as err:
+        logger.error("Error computing Euclidean distance: %s", err)
+        raise
+
+
+def manhattan(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray
+) -> float:
+    """Compute Manhattan distance between two embedding vectors.
+
+    :param embedding_1: First embedding vector.
+    :type embedding_1: np.ndarray
+    :param embedding_2: Second embedding vector.
+    :type embedding_2: np.ndarray
+    :return: Manhattan distance.
+    :rtype: float
+    """
+    try:
+        distance = np.sum(np.abs(embedding_1 - embedding_2))
+        return distance
+    except Exception as err:
+        logger.error("Error computing Manhattan distance: %s", err)
+        raise
+
+
+def jaccard(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray
+) -> float:
+    """Compute Jaccard similarity between two embedding vectors.
+
+    :param embedding_1: First embedding vector.
+    :type embedding_1: np.ndarray
+    :param embedding_2: Second embedding vector.
+    :type embedding_2: np.ndarray
+    :return: Jaccard similarity score.
+    :rtype: float
+    """
+    try:
+        intersection = np.minimum(embedding_1, embedding_2).sum()
+        union = np.maximum(embedding_1, embedding_2).sum()
+        similarity = intersection / union
+        return similarity
+    except Exception as err:
+        logger.error("Error computing Jaccard similarity: %s", err)
+        raise
+
+
+def pearson(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray
+) -> float:
+    """Compute Pearson correlation between two embedding vectors.
+
+    :param embedding_1: First embedding vector.
+    :type embedding_1: np.ndarray
+    :param embedding_2: Second embedding vector.
+    :type embedding_2: np.ndarray
+    :return: Pearson correlation coefficient.
+    :rtype: float
+    """
+    try:
+        correlation = np.corrcoef(embedding_1, embedding_2)[0, 1]
+        return correlation
+    except Exception as err:
+        logger.error("Error computing Pearson correlation: %s", err)
+        raise
+
+
+def minkowski(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray,
+    p: int = 3
+) -> float:
+    """Compute Minkowski distance between two embedding vectors.
+
+    :param embedding_1: First embedding vector.
+    :type embedding_1: np.ndarray
+    :param embedding_2: Second embedding vector.
+    :type embedding_2: np.ndarray
+    :param p: Power parameter for Minkowski distance (default is 3).
+    :type p: int
+    :return: Minkowski distance.
+    :rtype: float
+    """
+    try:
+        distance = np.sum(np.abs(embedding_1 - embedding_2) ** p) ** (1 / p)
+        return distance
+    except Exception as err:
+        logger.error("Error computing Minkowski distance: %s", err)
+        raise
+
+
+def hamming(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray
+) -> float:
+    """Compute Hamming distance between two embedding vectors.
+
+    :param embedding_1: First embedding vector.
+    :type embedding_1: np.ndarray
+    :param embedding_2: Second embedding vector.
+    :type embedding_2: np.ndarray
+    :return: Hamming distance.
+    :rtype: float
+    """
+    try:
+        if embedding_1.shape != embedding_2.shape:
+            raise ValueError(
+                "Embeddings must have the same length for Hamming distance."
+            )
+        distance = np.mean(embedding_1 != embedding_2)
+        return distance
+    except Exception as err:
+        logger.error("Error computing Hamming distance: %s", err)
+        raise
+
+
+def kl_divergence(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray
+) -> float:
+    """Compute Kullback-Leibler divergence between two probability 
+    distributions.
+
+    :param embedding_1: First probability distribution.
+    :param embedding_2: Second probability distribution.
+    :return: KL divergence.
+    """
+    try:
+        embedding_1 = embedding_1 / np.sum(embedding_1)
+        embedding_2 = embedding_2 / np.sum(embedding_2)
+
+        # Avoid division by zero and log(0)
+        embedding_1 = np.clip(embedding_1, 1e-10, 1)
+        embedding_2 = np.clip(embedding_2, 1e-10, 1)
+        divergence = np.sum(embedding_1 * np.log(embedding_1 / embedding_2))
+        return divergence
+    except Exception as err:
+        logger.error("Error computing KL divergence: %s", err)
+        raise
+
+
+def chebyshev(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray
+) -> float:
+    """Compute Chebyshev distance between two embedding vectors.
+
+    :param embedding_1: First embedding vector.
+    :type embedding_1: np.ndarray
+    :param embedding_2: Second embedding vector.
+    :type embedding_2: np.ndarray
+    :return: Chebyshev distance.
+    :rtype: float
+    """
+    try:
+        distance = np.max(np.abs(embedding_1 - embedding_2))
+        return distance
+    except Exception as err:
+        logger.error("Error computing Chebyshev distance: %s", err)
+        raise
+
+
+def bregman(
+    embedding_1: np.ndarray,
+    embedding_2: np.ndarray,
+    f=np.square,
+    grad_f=lambda x: 2 * x
+) -> float:
+    """Compute Bregman divergence between two embedding vectors using a convex
+    function.
+
+    :param embedding_1: First embedding vector.
+    :param embedding_2: Second embedding vector.
+    :param f: Convex function to compute divergence (default is square
+    function).
+    :param grad_f: Gradient of convex function. If not provided, defaults to 2
+    * x for f=x^2.
+    :return: Bregman divergence.
+    """
+    try:
+        divergence = f(embedding_1).sum() - f(embedding_2).sum() - \
+            np.dot(grad_f(embedding_2), (embedding_1 - embedding_2))
+        return divergence
+    except Exception as err:
+        logger.error("Error computing Bregman divergence: %s", err)
+        raise