Improvements to validation checks

[groutr]

Some minor improvements to _validation.py that center around the usage of np.asarray. Using np.asarray over np.array can avoid unnecessary copies (ie, when the input is already an array).

When comparing the shapes, we don't need to hold all the arrays in memory.

[aaraney]

@groutr, thanks for the contribution!

I am not sure the _array_attr function is necessary. np.asarray does not copy if it receives an input np.array. I think it would be more clear to just cast to np.asarray and check for the property in places you are proposing to call this method.

[groutr]

@aaraney _array_attr is there because these validation functions are most likely to receive pandas objects, not ndarrays. We are hoping to take advantage of pandas exposing the array properties without the unnecessary cast to ndarray. With _array_attr, checking the attribute is an order of magnitude faster.

# aa = pd.Series(range(1000))
In [93]: %timeit _array_attr(aa, "ndim").ndim
425 ns ± 41.8 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

In [94]: %timeit np.asarray(aa).ndim
4.45 µs ± 213 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

[aaraney]

Ah, I see. That makes a lot of sense. Thanks for clarifying and providing the above example, @groutr.

My only request would be adding some type hints to the function declaration. The return type hint might be difficult to nail down for all cases, so I think the straight path of npt.ArrayLike should be fine unless you have a better suggestion.

[jarq6c]

Looks good to me. I think we just want to bump the patch version in python/metrics/src/hydrotools/metrics/_version.py to 1.2.2

I tested these changes locally and it is indeed faster. Thanks for the contribution!

[aaraney]

Thanks again, @groutr!