You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
which corresponds to the definition given in the paper. Ditto for v_weight and e_reg_sum.
The text was updated successfully, but these errors were encountered:
yinyee
changed the title
Definition of e_weight as 1/w if w>0 else 1 leads to incorrect values of v_reg_sum
Definition of e_weight leads to incorrect values of v_reg_sumApr 29, 2021
In line 366, the values of
e_weight
are set to(1/w if w>0 else 1) for w in authorwt
. Why?Assuming that the
w
values inauthorwt
correspond to edge degrees, it seems to lead to an incorrect computation ofv_reg_sum
.Using this hypergraph as an example:
(Changing the indices from 1-based to 0-based)
Following line 366
Assuming
alpha_v = 1
, the results of line 382 should be:And
paper2sum
in line 383 should be:Then in line 393,
v_reg_sum
should be:But according to the definition of the normalisation denominator in the the paper:
the denominator matrix should be:
By setting the values of
e_weight
in line 366 tow
instead, we get:Then the results of line 382 should become:
And
paper2sum
in line 383 should become:Then in line 393,
v_reg_sum
should be:which corresponds to the definition given in the paper. Ditto for
v_weight
ande_reg_sum
.The text was updated successfully, but these errors were encountered: