Rotation in ellipsoid_plot function is incorrect. about ellipsoid_fit_python

[aleksandrbazhin]

Hi, thanks for feedback.
Actually, I borrowed this code from https://github.com/minillinim/ellipsoid - you can report there as well.
I tried to replicate bug by generating random sample on axis-aligned ellipsoid, before realizing that np.dot(rotation, [x[i, j], y[i, j], z[i, j]]) (proposed) and np.dot([x[i, j], y[i, j], z[i, j]], rotation) (current version) is exactly the same and is rotation[0]*x[i, j] + rotation[1]*y[i, j] + rotation[2]*z[i, j].
Could you please attach the input that is causing the error? There sure are bugs in that repository, I will try to fix them later.

[AlanKuurstra]

According to the matlab code (https://www.mathworks.com/matlabcentral/fileexchange/24693-ellipsoid-fit), evecs "contains the radii directions as columns of the 3x3 matrix". Therefore, in order for the evecs matrix to rotate a vector [1,0,0] on the unit sphere to the correct location on the ellipsoid, one should use:
np.dot(evecs,[1,0,0]) right?

Maybe I'm missing something.

I'll try to create an example if I find some time.

[aleksandrbazhin]

As for my previous answer - I was mistaken, of course rotation is a matrix. But opposed to Matlab code here evecs and rotation contain radii directions as rows not columns. I don't remember why I did it, probably to be able to extract ith eigenvector with evecs[i]. Maybe it was not a good idea as it may cause confusion.
Here I transpose eigenvectors before returning

ellipsoid_fit_python/ellipsoid_fit.py

Line 143 in 331dfb3

evecs = evecs.T

[AlanKuurstra]

Oh okay, I see. I missed the transpose when I glanced over the code.
If that's the case, then the plotting should actually be fine and this issue can be closed.