Test failures

[richard-lane]

Since #20, the tests are failing. I think this is because the definition of the contours' handedness/transforms/etc. have changed

[richard-lane]

I've had a look at some of the tests, and these might be the right fixes:

Closed vs open contours

I've had a quick look at the test failures, and I think some of them are due to inconsistencies with the contour's start and end point. contour_1 in the test file and simple_polygon in test_reconstruct_simple_contour are both closed contours (the start point is the same as the end point), but it looks like the new elliptic_fourier_descriptors function expects an open contour. Passing it a closed contour will set dxy[-1] to 0, which will cause all the coefficients to come out as NaN. Maybe the function should check that the contour is open; we could check for this with something like:

if contour[0] == contour[-1]: raise ValueError("Contour should be open...")

Normalising by size

When normalising, taking the absolute value of size might result in the coeffs coming out as [-1, 0, 0, ...] instead of [1, 0, 0, ...]. Do we want this

Other tests

The other tests for the transforms and reconstructing the contours are also broken - I haven't had time to look at these, but hopefully the above two things are helpful...

[hbldh]

I made some changes, one pure normalisation, one uncommenting and one is a comparison that I am not sure is mathematically solid, but it works code-wise...

Thank you for the looking through!

[hbldh]

@richard-lane I made some mergin, documentation in Changelog and created a new release, 1.7.0. It shoudl be present on PyPI to pip install now.