Fix reg()

[wleoncio]

This should close issue #28.

The translation of the original R definition of reg() was done in commit d4e3ef1, but that version incorrectly assigns the slope of new1 instead of the intercept to beta01(e). This is fixed here.

[wleoncio]

BTW, even without #17 being implemented on main, it seems like the C++ version is much faster than the R one. For legacy = FALSE I got an elapsed time of 34 s versus 111 s on the legacy code. Do you also get that?

[Theo-qua]

I agree with you that the c++ seems faster than the original R now. However, when I try with the huge X matrix, it seems to be slower. I checked with this

system.time( fit<-MADMMplasso(X,Z,y,alpha=alpha,my_lambda=NULL,lambda_min=0.01,max_it=5000,e.abs=e.abs,e.rel=e.rel,maxgrid=50,nlambda = nlambda, rho=5,tree = TT,my_print = F,alph=1,parallel =F,pal=1,gg=gg1,tol=tol,cl=6) )

instead of

system.time( fit<-MADMMplasso(X=X[,c(1:100)],Z,y,alpha=alpha,my_lambda=NULL,lambda_min=0.01,max_it=5000,e.abs=e.abs,e.rel=e.rel,maxgrid=50,nlambda = nlambda, rho=5,tree = TT,my_print = F,alph=1,parallel =F,pal=1,gg=gg1,tol=tol,cl=6) )