Sammendrag
It has always been a challenge to find good functional approximations to a set of observed data. Often the analyst uses linear models fitted to the data or to transformed data. This may work well for prediction purposes, but is less valuable for revealing the true nature of the phenomenon. The underlying model may be very complex. Estimation of complex model parameters may be time consuming, difficult and even impossible with standard statistical estimation methods. Another problem in statistical modeling is that the set of models explored typically is quite narrow. The choice of function types, which is made by the experimentalist, is usually quite subjective and according to his or her preferences and training. We seek a more objective approach to data modeling which considers a wide range of model classes and which has the capability of fitting models with minimal computation time. Therefore, the idea of having a library of possible curve shapes arose. This library can serve as a look-up database to get suggestions of the most plausible functions and parameter values that could form the given observed data points or data curves. This would help to reduce subjectivity and to speed up parameter estimation. In this work I present a first version of such a library and illustrate the approach on a set of 2D gel data for protein separation.
Vis fullstendig beskrivelse
