KneeLocator fails if there are flat extrema about kneed HOT 4 CLOSED

Thanks for posting the issue with an example @tomas-teijeiro, I'll take a look.

from kneed.

Great suggestion @tomas-teijeiro! See #53, using np.greater_equal and np.less_equal seems to address this, while also not breaking any of the other tests. It seems like this issue could be encountered frequently, so I'm going to push these changes to PyPI soon unless you can spot any potential errors.

k = KneeLocator(x, y, curve='convex', direction='decreasing', S=0.0)
k.plot_knee_normalized();

k = KneeLocator(x, y, curve='convex', direction='decreasing', S=1.0)
k.plot_knee_normalized();

from kneed.

Great! The only problem I can imagine with using np.greater_equal and np.less_equal is that if the difference curve has a staircase shape, then every step would be detected as a local maximum/minimum. However, since the algorithm selects the global one, I don't think it has any effect on the final result.

from kneed.

Agreed, that would be expected behavior, thanks again!

from kneed.