The simplify() method in shapely can be used to get rid of points that may be unncesesary for representing the object i.e. return a simplified version of the geometry object. See this link for further information on how the method works and is implemented. Sometimes it is unable to give me an accurate 'simplified' representation that I was looking for and I've found myself in heuristic battles playing with the tolerance values and using the Douglas-Peucker algorithm flags to get the right result.
More specifically I may come across a polygon that is represented by more coordinates than necessary (but has the correct number of vertices). Typically we only need the vertices to represent a simple polygon but sometimes there will be points along, and collinear, with the edges between vertices and simplify() fails to remove them.
Here I look at a method that uses simple trigonometry to look at the interior angles to eliminate these points
A generated hexagon with 6 vertices (technically 7 coordinate points for a complete linear ring sequence, but not important!). Consider this the goal result.
now I'll add randomly generated points along the edges to emulate the types of problematic polygons mentioned earlier (here I have randomly generated 5 points per side to be collinear with the vector between vertices)
Below is an attempt using the simplify() method with preserve_topology = False (which corersponds to using the Douglas-Peucker algorithm) and a range of tolerance values: [1e-10, 0.1, 1, 10, 100] (left to right, respectively)
As you can see there is still a persistent 'non vertex' point in this example. And at some point (i.e. tolerance value of 10) the structure of the hexagon is lost. At 100 it is gone!
Same approach as above, but with preserve_topology = True
Same deal, persistent points but in this case at a tolerance of 100 a polygon still exists.
Now if one runs the method of reducing points via interior angles as proposed we get...
as desired!
This hasn't been rigorously tested. As you can imagine it hasn't been straight forward in pre-determining when the existing simplify method wouldn't work on a given polygon. It is also a worthwhile reminder that the existing simplify() method has a lot more uses than what is presented here (eg. simplifying line-strings, dimensionality reduction of cartographic objects, etc.). This is potentially more of a niche problem.