If it can be coded, I prefer having both, or implementing the code. It helps in understanding the algorithm behind. But maths is much larger than what can be coded, or is useful in code, so the only thing left is playing with toy examples ("coding" when working with really weird stuff.)

I'd love to see more of APL (and a "larger" set of APL functions, actually) in use. The idea of a notation we could run directly is/was awesome.