I'm not sure how good that would be for helping you understand the formulas. I think they would probably make them more obscure, especially with optimizations. What if you tried to teach yourself Linear Algebra just by reading the source code of Intel's MKL (if you had access to it); I don't think you'd get as far as if you just studied a traditional math based approach.

Absolutely. Although as a programmer and someone who reads mathematical papers I'm not sure how much of a benefit this would be. The hard part in understanding isn't any particular step, its comprehending how each step fits together as a whole. I don't immediately see how the translation would help this.

I just feel like I think a lot better looking at code. Each input would have a descriptive variable name.

And sometimes if a formula has something unusual like a derivative, it's hard for me to see how that converts to code.

I feel the same way. In fact, reading math feels like reading code where all the operators have been overloaded and there are only 26 available variable names.

Sounds like the math you're reading needs better Unicode support :)

Yeah, there's no doubt a benefit in reading a concept in a language you're more comfortable with. It would have to be an interesting algorithm that could extract meaningful variable names from the surrounding prose when translating an equation to code :)

I think the NumPy people are the best team for making this. We should lobby them!