Methods for computer self-organization of models of real objects and processes based on observational data under interference conditions are considered. The main attention is paid to effective relaxation-iterative algorithms that allow solving modeling problems with hundreds of thousands of variables. The most rapid algorithm is described, which, unlike others, has linear computational complexity at the stage of model building. The problem of sample partitioning and the coordination of the criteria of external addition with the criteria for sample splitting in the method of group accounting of arguments are studied. For specialists in the field of inductive modeling of complex systems of diverse nature.