PERF: improve MultiPoint(..) constructor from numpy array of coordinate values

[jorisvandenbossche]

Addresses #1954

[coveralls]

Pull Request Test Coverage Report for Build 7401687526

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.3%) to 88.01%

---

Totals
Change from base Build 7178707979:	0.3%
Covered Lines:	2591
Relevant Lines:	2944

---

💛 - Coveralls

[jobh]

Looks good!

Just a heads-up, there is a semantic change because an all-NaN point is considered an empty geometry:

coords = np.array([[np.nan, np.nan]])
shapely.MultiPoint(coords)  # before PR
# EmptyPartError: Can't create MultiPoint with empty component
shapely.MultiPoint(coords)  # after PR
# <MULTIPOINT EMPTY>

[jorisvandenbossche]

Good point! We should try to keep that consistent. The points() constructor now has a handle_nan keyword that can be set to disallow NaNs.

Now, that would also disallow single NaNs for just x or y, which currently is allowed:

In [7]: shapely.MultiPoint([(0, 1), (2, np.nan)])
Out[7]: <MULTIPOINT ((0 1), (2 NaN))>

so if we want to retain the actual current behaviour of MultiPoint(..), we could also first create points, and then just use the vectorized method to check if there are any empties in there.