start with a trial density, say
with weighted Monte Carlo samples, estimate the parameters, mean and covariate matrix, and construct new trial density, say
construct a certain measure of discrepancy between the trial distribution and the target distribution, such as the coefficient of variation of importance weights, does not improve any more.
Implement an Adaptive Importance Sampling algorithm to evaluate mean and variance of a density
A possible procedure is as follows:
start with a trial density
in which one chooses to minimize the variation of coefficient of the importance weights.