Improving convergence of the stochastic decomposition algorithm by using an efficient sampling technique