Some questions about the code

[HqChen2021]

hi,

Thanks for sharing your code. I'm following up on your work. I have some questions about the code.

1. what is self.Ca1 in class HCO_LP? I assume it is a variable representing the gradient? (I assume this from the self.prob_dom definition)
2. I don't get the point in line 71

   FCFL/FUEL/hco_lp.py

   Line 71 in 5930200

   self.Ca1.value = (d_ls @ d_l1.t()).cpu().numpy()

   , self.Ca1.value = (d_ls @ d_l1.t()), It seems d_ls is the grad_l (gradient produced by the classification loss) and d_l1 is grad_g (gradient produced by the fairness-related loss).
3. In the same class, line 24

   FCFL/FUEL/hco_lp.py

   Line 24 in 5930200

   constraints_dom = [self.alpha >= 0, cp.sum(self.alpha) == 1]

   , are there some constraints missing?

[zaocan666]

d_ls=[grad_l, grad_g]^T (line 68)
We want to compute the gradient d=alpha @ d_ls, so we only need to get the alpha.
when the fairness constraint is satisfied, optimization goal is alpha @ d_ls @ d_ls^T, it is consistant with the paper.