viggin / domain-adaptation-toolbox Goto Github PK

New website of SSA is
https://mloss.org/software/view/292/

Hi, I have a question about the code in line 23th~26th in domain-adaptation-toolbox/ToRelease_GFK/GFK.m, that is,

B1 = 0.5.diag(1+sin(2theta)./2./max(theta,eps));
B2 = 0.5.diag((-1+cos(2theta))./2./max(theta,eps));
B3 = B2;
B4 = 0.5.diag(1-sin(2theta)./2./max(theta,eps));

I checked the corresponding formula (Eq. 6) in the article "Geodesic Flow Kernel for Unsupervised Domain Adaptation" and found that there was no factor 0.5 in the formula.

I sincerely hope you can spend some time to answer my doubts. Thanks.

Hi, I have a question about the code in line 27th~29th in domain-adaptation-toolbox/infometric_0.1/compute_mutual_infoL.m, that is,

line 27: P = expD ./ repmat( sum(expD), Ns, 1);
line 29: P = P - diag( diag(P) ).

In fact, these two lines of code are inconsistent with equation 2 in the paper "Information-Theoretical Learning of Discriminative Clusters for Unsupervised Domain Adaptation".
When normalizing the exponential distance matrix (expD), the contribution of the element to the denominator when the row index (i) is equal to the column index (j) should not be taken into account.

I sincerely hope you can spend some time to answer my doubts. Thanks.