Hi Ce Qin, What you are doing will work fine. There is no back-tracking in a linesearch to a previously computed x, the last x computed is the current solution.
Jason Sarich On Sun, May 25, 2014 at 7:11 AM, Ce Qin <[email protected]> wrote: > Hi Barry, > > I think there is confusion about my question. > > The purpose of line search is to find an \alpha_{k} that minimize f(x + > \alpha p). This procedure usually choose several \alpha and returns a > proper \alpha_{k}. For each \alpha, we need to compute the objective > function and gradient of (x + \alpha p). In my FormFunctionGradient > function, I save some internal data(model response) which will be printed > in the monitor routine. If the objective function and gradient returned by > TaoLineSearchApply isn't the latest computed, my internal data becomes > invalid in the monitor routine. My question is the order of the objective > function and gradient corresponding to \alpha_{k} in this line search > procedure. > > If it is not clear, please let me know. Thanks. > > Best regards, > Ce Qin >
