Stretching matrices

[oleg-kachan]

Dear Pedro,

Can you explain lines 177, 180, 184 in utils.py?

p = np.sqrt(disp_target / disp_source) target_pow_train['covs'] = np.stack([powm(covi, 1.0/p) for covi in covs_target_train]) target_pow_test['covs'] = np.stack([powm(covi, 1.0/p) for covi in covs_target_test])

You are taking powers of 1/p instead of p, it seems correct because you are computing p as the fraction of the target over source instead of the source over target as shown in paper eq. 16, but why?

[plcrodrigues]

Hi, the goal with the stretching operation is to ensure that the dispersion for target is the same as that for source. Of course, I could have done things with p = np.sqrt(dist_source/disp_target) and then used p in the powm operation... it would be simpler and easier to understand. However, it I remember correctly, at the time when I wrote this code, I had other pipelines going on where the inverted-p operation was more natural. Anyways, it's just a matter of taste in the end :-)