Internal function. With stratified samples, calculate the variance of the estimate from importance sampling without control variates
1  get.var(Y, nvec)

Y 
vector of stratified samples of length n. i.e. Y_1 = Y[1:nvec[1]] are sampled from q_1, Y_i = Y[(nvec[i1]+1):nvec[i]] are sample from q_i. 
nvec 
the vector of number of samples from each mixture component. It sums up to n. 
Suppose we sample Y from a mixture q_{α} = α_1*q_1 + ... + α_J*q_J. To estimate \mathrm{mean}(Y), fixing the number of samples from each mixture component and getting a stratified sample would reduce the variance of the estimate. The formula for \mathrm{Var}(\hat{μ}) with stratified samples is
\mathrm{Var}(\hat{μ}) = 1/n \times ∑_{j=1}^J α_j \mathrm{Var}(Y_j)
the variance estimate of \hat{μ} = 1/n ∑_{i=1}^n Y[i]
Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.
All documentation is copyright its authors; we didn't write any of that.