Tuesday, August 21, 2012

Improving the Solver

I am working on improving the Solver and investigating ways to replace (at least in some cases) the current algorithm which is based on the BFGS method with an algorithm based on the Gauss-Newton algorithm which should be easier to compute.

I'm also updating the error functions to simplify the calculations and adding the calculated values for the first derivative for each function: each of the point's coordinate is a variable for the Jacobian matrix and for each method we'll have several expressions to evaluate depending on the coordinate. I'll explain this in more detail when the algorithm will be finalized.

I've also added two helper classes for vector and matrix calculations - these should make the algorithm a lot easier to understand and make debugging a lot easier.

No comments: