optimizealpha does not converge for alphas >= 4

[Dapid]

• Alpha Shape Toolbox version: 1.1
• Python version: 3.7
• Operating System: Fedora 32

Description

optimizealpha fails when the optimal alpha is a bit big. In my particular case, upper and lower values are:

lower = 4.5 
upper = np.nextafter(lower, 9)
print((upper - lower) > numpy.finfo(float).eps * 2 )         
True                                                                                                                           

The middle point is identical to lower, so the model never converges below the threshold.

Solutions:

I see two ways of solving the problem:

• Compute the right eps for the current value range: 2 * np.finfo(float).eps * lower, or more accurately, np.finfo(float).eps * 2 ** math.floor(1 + math.log2(lower)),
• Allow the user to specify their own threshold.

The first case is better suited to the current code, but getting alpha to within machine precision may not always be necessary.

[jazzathoth]

I am having similar trouble with latitude longitude data.