Cosine similarity with nan checks

cosine_similarity.py

from numpy.linalg import norm
import numpy as np

# Define the cosine similarity function with automatic list-to-array conversion
def cos_sim(a, b):
    # Check if either input is NaN, empty, or contains empty strings
    if a is None or b is None or not a or not b:
        return np.nan
    if isinstance(a, list) and any(x == "" or x is None for x in a):
        return np.nan
    if isinstance(b, list) and any(x == "" or x is None for x in b):
        return np.nan
    
    # Convert lists to NumPy arrays if they are not already
    if isinstance(a, list):
        a = np.array(a)
    if isinstance(b, list):
        b = np.array(b)
    
    # Check if the arrays contain NaN values
    if np.isnan(a).any() or np.isnan(b).any():
        return np.nan
    
    # Calculate cosine similarity
    return np.dot(a, b) / (norm(a) * norm(b))

# Example usage
print(cos_sim([1, 2, 3], [4, 5, 6]))  # Should return a valid cosine similarity value
print(cos_sim([np.nan, 2, 3], [4, 5, 6]))  # Should return np.nan
print(cos_sim([], [4, 5, 6]))  # Should return np.nan
print(cos_sim(None, [4, 5, 6]))  # Should return np.nan
print(cos_sim(["", 2, 3], [4, 5, 6]))  # Should return np.nan