I have two sets of vectors A,B and I want to compute the TS-SS similarity for each vector in A compared to every vector in B.
I have an implementation (from https://github.com/taki0112/Vector_Similarity) used for 2 vectors, however when I tried using it for matrices (function "compute_matrix_sim") - it is very inefficient and takes forever.
Is there a more vectorized approach to use TS-SS for matrices?
Thanks!
import math
import numpy as np
class TS_SS:
def Cosine(self, vec1: np.ndarray, vec2: np.ndarray):
return np.dot(vec1, vec2.T) / (np.linalg.norm(vec1) * np.linalg.norm(vec2))
def VectorSize(self, vec: np.ndarray):
return np.linalg.norm(vec)
def Euclidean(self, vec1: np.ndarray, vec2: np.ndarray):
return np.linalg.norm(vec1 - vec2)
def Theta(self, vec1: np.ndarray, vec2: np.ndarray):
return np.arccos(self.Cosine(vec1, vec2)) + np.radians(10)
def Triangle(self, vec1: np.ndarray, vec2: np.ndarray):
theta = np.radians(self.Theta(vec1, vec2))
return (self.VectorSize(vec1) * self.VectorSize(vec2) * np.sin(theta)) / 2
def Magnitude_Difference(self, vec1: np.ndarray, vec2: np.ndarray):
return abs(self.VectorSize(vec1) - self.VectorSize(vec2))
def Sector(self, vec1: np.ndarray, vec2: np.ndarray):
ED = self.Euclidean(vec1, vec2)
MD = self.Magnitude_Difference(vec1, vec2)
theta = self.Theta(vec1, vec2)
return math.pi * (ED + MD) ** 2 * theta / 360
def __call__(self, vec1: np.ndarray, vec2: np.ndarray):
return self.Triangle(vec1, vec2) * self.Sector(vec1, vec2)
def compute_matrix_sim(m1, m2):
similarity = TS_SS()
ans = np.zeros((len(m1), len(m2)))
for i, vec_m1 in enumerate(m1):
for j, vec_m2 in enumerate(m2):
ans[i, j] = similarity(vec_m1, vec_m2)
return ans
# Usage
v1 = np.random.random_sample((40, 80))
v2 = np.random.random_sample((2000000, 80))
x = compute_matrix_sim(v1, v2)
print(x)