[Help-gsl] fitting example

[Petr Ent]

hi,
i have next (hopefully last :-) ) question about non-linear fitting example. 
function 
    
int expb_f (const gsl_vector * x, void *data,  gsl_vector * f)
"this function should store the vector result f(x,params) in f for argument x 
and arbitrary parameters params"

i understand it that you should compute the value of your function to be fitted 
with current results (x vector) and store it to vector f. but in this example, 
you do that + substract "real value" (y_i) from it and divide it by deviation 
(sigma_i).

gsl_vector_set (f, i, (Yi - y[i])/sigma[i]);

why do you do that? and what if you dont know deviation of data to be fitted?

thanks for answering
Petr Ent