Minimize useless tweaking of a numeric app
- by Potatoswatter
I'm developing a numeric application (nonlinear optimizer), with a zillion knobs to tweak and rising.
It's not my first foray into this domain, but this time there are even more variables in the code and I'm on a tight schedule. Don't want to waste time fiddling.
Days or even months can potentially be wasted adjusting variables, recompiling, and reprocessing benchmark datasets. The resulting data is viewed and trouble spots are checked. The overall quality of the solution is reported by the program but the meaning of the report could change over time. (Numeric units for the report are one thing I'm trying to nail down.)
One main problem is organizing result files to identify each with specific code changes. Note taking can be a pain, is there software to help with this?
Are there agreed best practices to making this kind of development cycle reliably move forward? The solver package converges to its optimal solution with mechanical determination, but I'm all too familiar with the way an excess of design decisions can mire development.
