This is an extension of an earlier question asked here. I have a data-set coming from a distribution given by:
$$Pr(X=k)=p \left({n\choose k}a^k(1-a)^{n-k}\right)+(1-p) \left({n\choose k}b^k(1-b)^{n-k}\right)$$
which is a mixture of two binomial distributions.
My aim is to estimate the parameter values, namely: n, p, a, b. To this I want to optimize the maximum likelihood function using "JDEoptim" package via R software.
fn<-function(x){
#x(1)=p; x(2)=n; x(3)=a; x(4)=b;
-prod(DensityFn(x, dataset[1:length(dataset)]))
}
DensityFn<-function(x, val){
x[1]*dbinom(val, floor(x[2]), x[3])+(1-x[1])*dbinom(val, floor(x[2]), x[4])
}
lower=c(0, 11, 0, 0);
upper=c(0.5, 20, 1, 1);
print(JDEoptim(lower, upper, fn))
The variable "data-set" contains the values generated by this density function. The output upon running this program gives me:
$par [1] 0.07007818 11.85238114 0.86411002 0.64941201
$value [1] 0
$iter [1] 0
$convergence [1] 0
'iter' is 0 while 'convergence' is 0...indicating convergence has occured, however the parameter values outputted are far off from the actual values. Also, every time I run the program, the parameter estimates differ, but 'iter' and 'convergence' are always 0.
Is my code and logic correct?
