Optimization Suggestion: Replace np.sum(A) with A.sum() for better performance

[SaFE-APIOpt]

pyod/pyod/models/sos.py

Line 29 in d595486

sumA = np.sum(A)

Hi, I’d like to suggest a small performance improvement in this code:

A = np.exp(-D * beta)
sumA = np.sum(A)

This can be more efficiently written as:

A = np.exp(-D * beta)
sumA = A.sum()

Since A is the direct result of a NumPy ufunc (np.exp), it is guaranteed to be a NumPy ndarray. When using np.sum(A), NumPy performs a layer of general-purpose dispatching, including checking the input type, handling duck-typed objects, and possibly resolving array_function overrides. Even though these checks are redundant here, they still introduce minor overhead. In contrast, A.sum() directly invokes the internal C implementation associated with the array object itself, bypassing unnecessary logic and executing with minimal function call depth. This makes it the preferred choice in performance-sensitive contexts where the input type is known.