General Modeling Library


Short Description

The software in what I am currently calling the General Modeling Library (GML for short), is also built upon other people's work. Several core functions (class methods) are modifications of code originally written by William McIllhagga. This includes a version of Powell's method for fitting of functions and a method for finding rigorous 95% (or other) confidence intervals that proceeds by searching the error surface. You need only supply the surface-level-increment for which to search. Powell's method and the rigorous confidence interval algorithm can be found by perusing the Numerical Recipes in C tome, but the routines contained here were written directly and therefore require no permission for distribution.

Also included is an array of goodness-of-fit measures taken from various sources (Coefficient of Determination, r-squared, Model Selection Criterion), as well as code for calculating the error surface Hessian, and from the Hessian, parameter covariance and correlation matrices. The covariance matrix allows a less rigorous but often more useful method for calculating confidence intervals, as well as means by which to analyze the utility of the parameters in capturing the data fit by the model. This is achieved through use of actual Numerical Recipes in C functions, so you need to acquire a copy of NRC to employ them.

You should be able to compile and use GML with other compilers than the one from Metrowerks. More importantly, this software should run on any architecture, including Mac, PC/Intel or your favourite Unix box. You just need to get the NRC routines onto the platform. I know so because the code has been developed over time on both PC and Mac. :-)

General Overview

Useful Definitions

Detailed Documentation