Enhancement wishlist

[ljwolf]

This is a running list of enhancements burbling up of when writing the gdsbook/book chapter on point pattern analysis.

additional hulling measures

• hull() should be general, with a type argument that supports alpha shapes, minimum bounding rectangles, minimum area rectangles, and minimum bounding circles. This would require putting a dedicated convex_hull function elsewhere. Since we could set hull's default to be the convex hull, this can actually be made API-transparent :smile:
• mbr should be expanded to minimum_bounding_rectangle
• we should make an import-safe minimum_area_rectangle and borrow/import the opencv implementation of the minimum area rectangle.
• we should rename skyum to minimum_bounding_circle, or use it as a fallback if we cannot use opencv's minimum bounding circle
• docstring for skyum needs to be brought in line with other functions/classes (my bad :disappointed_relieved:)

API consistency

• Elsewhere in the library, we avoid exposing everything at root now. Here, though, everything in centrography is available directly in pointpats... is this intentional?
• skyum should return in the same manner as ellipse. I think they both should give some kind of namedtuple return value that has (center, radius) or (center, semimajor, semiminor, rotation).

Performance

• mbr does a loop in python through all points and explicitly finds the minimum. We should probably use a sort for a faster solution?
• skyum can be easily numba-ized, but I'm not sure if it'd matter much.

New statistics

• local K/F functions

[sugam45]

Hi @ljwolf I have analyzed the points put forth by you. In my analysis :

• A generic hull function is a really good suggestion. This can be efficiently done via a generic hull class importing all other functions.

• I agree with the import safe suggestion. This will really help in removing breakpoints in our code.

• Sort for mbr may not be a prudent choice (sort being a O(N log N) ). Instead, we can use a comparison in pairs strategy.

• We can have a namedtuple each place functions are getting referenced.

• I think that numba-ization can be done for skyum. We can do it via parallel threading

[ljwolf]

Hey thanks! If you want to take a stab at it,

1. wrapping the opencv minimum area rectangle into a special function
2. renaming mbr to minimum bounding rectangle

would be a good first steps for this!

[sugam45]

Hi @ljwolf I have already made some changes in the code for mbr loop.

• Instead of a loop, the tuple array has been divided into 2 arrays and separate max and min have been calculated.

• Also, I have renamed mbr to minimum_bounding_rectangle.

• Currently, I am working on the wrapper for opencv minimum area rectangle.

I will raise the PR for the same after proper testing.