bundles_distances_mdf asymmetric values

[bloomdt-uw]

Description

Using bundles_distances_mdf to calculate minimum average direct-flip distances between streamlines on subjects from HCP dataset results in asymmetric values. Expecting distance to be symmetric: MDF(i,j)=MDF(j,i)

>>> import numpy as np
>>> from dipy.io.streamline import load_tractogram
>>> import dipy.tracking.streamline as dts
>>> tractogram = load_tractogram(tg_fname, 'same')
>>> streamlines = tractogram.streamlines
>>> fgarray = np.array(dts.set_number_of_points(streamlines, 100))
>>> mdf = dts.bundles_distances_mdf(fgarray, fgarray)
>>> np.count_nonzero(mdf-mdf.T)
2930

Asymmetries are small, so this may be a result of numerical error:

>>> np.sum(abs(mdf-mdf.T))
0.014194488525390625

Way to reproduce

Code example

>>> import numpy as np
>>> import dipy.tracking.streamline as dts
>>> A = np.load('max_asymmetric_mdf.npy')
>>> dts.bundles_distances_mdf(A,A)
array([[ 0.        , 27.01378632],
       [27.01377106,  0.        ]])

max_asymmetric_mdf.npy.zip

Relevant images

Operating system and version

macOS-10.16-x86_64-i386-64bit

Python version

Python 3.8.5 (default, Sep  4 2020, 02:22:02) 
[Clang 10.0.0 ]

dipy version

dipy 1.2.0

dependency version

NumPy 1.17.5

[skoudoro]

Hi @bloomdt-uw,

Thank you for this nice report. Can you look into this @BramshQamar and @Garyfallidis? Thanks!

[skoudoro]

Sorry for the late reply @bloomdt-uw.

I just had a look and fixed it on #2389. It should be good for the next release in June