MATLAB: Can I perform D-optimal experimental design in conjuction with the ROWEXCH function in Statistics Toolbox 5.0 (R2006b)

Statistics and Machine Learning Toolbox

I am doing a D-optimal experimental design. My parameter space is limited. I need to augment the experiment space incrementally. I see the DAUGMENT function but am not sure how to use the two together.

Best Answer

There is no feature in CANDEXCH or ROWEXCH for augmenting the experiment space incrementally. The function DAUGMENT (for the coordinate exchange algorithm) is therefore not suited.
The attached example uses a modified version of CANDEXCH, which has been modified to support this idea.
The 'candexch.m' file has been modified to support a new ‘start’ argument, allowing you to specify an initial set of runs to be augmented. It needs to reside in the $MATLAB/toolbox/stats directory because it calls functions residing in $MATLAB/toolbox/stats/private (where $MATLAB is your MATLAB root directory).
The attached 'testcandexch.m' file is a little program used to test the results. It compares the row-exchange results from CANDEXCH with the coordinate-exchange results from DAUGMENT. The first few lines have parameters that you can adjust to try other test conditions.
Note that these functions have not passed our quality control tests. We can therefore not guarantee that they will work in all situations.
Related Question