Monte-Carlo Simulation of Error Sort Model

We propose a novel quantization method for human likelihood or preference using rank data. We estimate an evaluation function by assuming that the data is sorted on the basis of observed function values and that the observation includes some errors. We use the difference in the rank order to estimate the corresponding function. Then, its performance will vary depending on the settings. Our data are normally distributed random numbers. First, function values are given as normally distributed random numbers, and after that, observed values are given by adding normally distributed random errors. The performance will depend on the difference in the variances of two normal distributions: for the function values and for the error values. We present a test to evaluate the proposed method by using random data.