Absolute convergence and error thresholds in non-active adaptive sampling

Non-active adaptive sampling is a way of building machine learning models from training data base which are supposed to dynamically and automatically derive guaranteed sample size. In this context regardless the strategy used in both scheduling generating weak predictors, proposal for calculating absolute convergence error thresholds described. We not only make it possible establish when quality model no longer increases, but also supplies proximity condition estimate terms how close achieving such goal, thus supporting decision making fine-tuning parameters selection. The technique proves its correctness completeness with respect our working hypotheses, addition strengthening robustness scheme. Tests meet expectations illustrate domain natural language processing, taking generation part-of-speech taggers as case study.