Dear experts,
I have a question about the permutation strategy in the two-sample discriminability test. When comparing two Discr estimates, I noticed that the current implementation in discrim_two_sample.py uses a different permutation approach than described in the original paper (Bridgeford et al., 2021).
Current implementation (lines 190-191 in discrim_two_sample.py):
permx1 = self._get_convex_comb(self.x1, random_state) # convex combination within x1
permx2 = self._get_convex_comb(self.x2, random_state) # convex combination within x2
Paper description (Bridgeford et al., 2021):
The paper suggests creating "randomly combined datasets" by taking "random convex combinations of the observed data from each of the two methods choices", which I interpret as mixing data between the two input matrices.
Could you help me understand the rationale behind performing convex combinations within each matrix rather than between them? Is this an intentional design choice for specific theoretical or practical reasons?
Thank you for your time and clarification!
Best regards,
Alex
Dear experts,
I have a question about the permutation strategy in the two-sample discriminability test. When comparing two Discr estimates, I noticed that the current implementation in
discrim_two_sample.pyuses a different permutation approach than described in the original paper (Bridgeford et al., 2021).Current implementation (lines 190-191 in
discrim_two_sample.py):Paper description (Bridgeford et al., 2021):
The paper suggests creating "randomly combined datasets" by taking "random convex combinations of the observed data from each of the two methods choices", which I interpret as mixing data between the two input matrices.
Could you help me understand the rationale behind performing convex combinations within each matrix rather than between them? Is this an intentional design choice for specific theoretical or practical reasons?
Thank you for your time and clarification!
Best regards,
Alex