Question

We have not been able to pinpoint what is causing the error of Invalid Function Value in Optimize in our Optimizing code. If you could offer any insight, it would be appreciated.

H_fun <- function(c) 
{ 
val = -current_c_weight*c - X_counts%*%log( 
exp(rep(c,length(current_Theta))*current_Theta) - 
current_elongation_rates ) 
print('#########iteration display#############') 
print('c') 
print(c) 
print('val') 
print(val) 
print('current_c_weight') 
print(current_c_weight) 
print('current_Theta') 
print(current_Theta) 
print('current_elongation_rates') 
print(current_elongation_rates) 
} 

#...snip...

# minimize -H(c) without the non-negativity constraint 
#tmp = optim(c(0,1),H_fun,NULL, method = "BFGS", hessian = TRUE); 
tmp = optimize(H_fun,interval = c(0,1)); 

Here is a link to the code:

http://www.text-upload.com/read.php?id=102950&c=8605046

Answer 1

Are you sure H_fun is returning a one-dimensional value?

Look at fcn1() in the R optimize() source code:

static double fcn1(double x, struct callinfo *info)
{
    SEXP s;
    REAL(CADR(info->R_fcall))[0] = x;
    s = eval(info->R_fcall, info->R_env);
    switch(TYPEOF(s)) {
    case INTSXP:
        if (length(s) != 1) goto badvalue;
        if (INTEGER(s)[0] == NA_INTEGER) {
            warning(_("NA replaced by maximum positive value"));
        return DBL_MAX;
        }
        else return INTEGER(s)[0];
        break;
    case REALSXP:
        if (length(s) != 1) goto badvalue;
        if (!R_FINITE(REAL(s)[0])) {
            warning(_("NA/Inf replaced by maximum positive value"));
            return DBL_MAX;
        }
        else return REAL(s)[0];
        break;
    default:
        goto badvalue;
    }
 badvalue:
    error(_("invalid function value in 'optimize'"));
    return 0;/* for -Wall */
}

goto badvalue occurs if length is not 1. Also, the package summary states that optimize() works on a one-dimensional unconstrained function.