Ok - I gave the genetic optimizer freedom to pick the y values of the spline knot points, I left the x values as logarithmically spaced from 1e-4 to 1e1. I ran a comparison between phi0 left the way that it was in the original source code and the changes introduced by the optimizer.
An optimizer step included rewriting the "phi0.c" file, compiling the project, running the test script, extracting PER for 1000 samples. ... Sadly, the spline fit is still merely faster, it has somewhat higher PER. I picked 1000 for the samples because it seemed to have reached asymptotic behavior by then. (I'm also guessing "faster" because more iterations were performed during the test? Perhaps I am wrong) I guess I will try linear interpolation and then nearest neighbor approximation ...
_______________________________________________ Freetel-codec2 mailing list Freetel-codec2@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/freetel-codec2
