SNOW-2048136: Optimization: Use arr.item(0) instead of arr.flat[0] for faster dtype detection

[SaFE-APIOpt]

snowflake-ml-python/snowflake/ml/data/torch_utils.py

Line 98 in dde003f

arr = np.array(array_list, dtype=arr.flat[0].dtype) if isinstance(arr.flat[0], np.ndarray) else array_list

Hi, I’d like to suggest a small performance optimization to the following expression:
arr = np.array(array_list, dtype=arr.flat[0].dtype) if isinstance(arr.flat[0], np.ndarray) else array_list
This can be simplified and made more efficient as:
arr = np.array(array_list, dtype=arr.item(0).dtype) if isinstance(arr.item(0), np.ndarray) else array_list
The use of arr.flat[0] introduces unnecessary overhead by first constructing a flatiter object and then performing index resolution through the iterator interface. In contrast, arr.item(0) is a direct call to a C-level function that extracts the scalar value at the specified flat index without intermediate object creation. It is faster, semantically cleaner, and returns a Python-native object that behaves more predictably in type checks and scalar operations. Since the goal here is simply to detect the type and extract the .dtype attribute from the first element, item(0) is the more appropriate and performant choice.

[sfc-gh-shchen]

Thanks for the good suggestion @dbsi-pinkman , we will look into it.