Group lasso with tfocs


#1

I am having problems implementing group lasso with TFOCS. The problem that i am solving is as follows.
\min_{a} \frac{1}{2} ||y-Xa||^2+\lambda\sum_{j=1}^N ||a_{j}||_2.
Here a is a vector of size Np, which can be thought of consisting of N groups, with each group having p elements, and the non-smooth penalty that I am trying to put is the standard group lasso.

The problem that I am facing is that when I call tfocs, with prox_l1l2 as the non-smooth function, then it seems to just use the L_1 norm of x, since x is a vector. In principle it seems like it should be possible to just reshape x, and pass it to prox_l1l2, but the only argument that prox_l1l2 can take is a scalar argument.

This is my first experience with tfocs, and any help would be deeply appreciated.