Optimal properties of the Bechhofer-Kulkarni Bernoulli selection procedure (Q1087261)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Optimal properties of the Bechhofer-Kulkarni Bernoulli selection procedure |
scientific article |
Statements
Optimal properties of the Bechhofer-Kulkarni Bernoulli selection procedure (English)
0 references
1986
0 references
The paper discusses sequential procedures for selecting the best among several Bernoulli populations. Let \(\pi_ i\) (1\(\leq i\leq k)\) denote \(k\geq 2\) Bernoulli populations with corresponding success probabilities \(p_ i\). Denote the ordered \(p_ i\) values by \(p_{[1]}\leq...\leq p_{[k]}\); let \(\bar p=(p_{[1]},...,p_{[k]})\). The goal of the experimenter is to select the population which corresponds to \(p_{[k]}\). When such a population is selected, we say a correct selection (CS) has been made. Define \(N_{(i)}\) to be the number of observations from the population associated with \(p_{[i]}\) at the termination of sampling and let \(N=\sum N_{(i)}\). Let \(E_{{\mathcal P}}(N| \bar p)\) denote the expected value of N under procedure \({\mathcal P}\) for a given value of \(\bar p.\) A class \({\mathfrak C}\) of procedures is defined which achieve the same P(CS), uniformly in \(\bar p,\) as does the single stage procedure which takes n observations from each population and selects the population with largest number of successes breaking ties at random. The procedures in \({\mathfrak C}\) share a common stopping rule and terminal decision rule but use different sampling rules; they take no more than n observations from any one population. Various results are proved for the case \(k=2\). For \(k=2\), a procedure \({\mathcal P}^*\) is discussed, which minimizes E(N\(| \bar p)\) among \({\mathfrak C}\) whenever \(p_{[1]}+p_{[2]}\geq 1\). A new procedure is described for the case \(k\geq 3\) and it is shown that the procedure minimizes the expected total number N when all the success probabilities are small.
0 references
Bechhofer-Kulkarni Bernoulli selection procedure
0 references
probability of correct selection
0 references
expected total number of observations
0 references
adaptive sampling
0 references
new selection procedure
0 references
sequential procedures for selecting the best among several Bernoulli populations
0 references
stopping rule
0 references
terminal decision rule
0 references