I'd like to respectfully ask for some advice on this forum. If you have to program complex mathematical code, do you prefer to write the code yourself or obtain an algorithm from a library of such downloadable code. I'm especially interested who has real experience in this area. Have you had good luck with the libraries?

Just for clarity sake, I mean a library like Numerical Algorithm Group (www.nag.com), that costs $3000 per year for unlimited downloads. The code is supposed to be checked out and they offer assistance implementing the code. Do you think this a bargain, have you ever tried such a service? Thanks for any advice you can share on this topic.

It really comes down to what $3000 means to you. Most of these purchased algorithms come with some guarantee of performance and accuracy, as well as you mentioned, the tech support and assistance being offered.

 

For my personal applications, I have always found that I could write my own or use freely available algorithms that performed well enough.

 

But, if performance, accuracy, or development time were all at a premium, it could very well be worth $3000 or more. If the given budget for a project is $750,000, $3000 may be meaningless. If the budget is $5000, $3000 is clearly mountainous. It also is worth being very clear what the licenses and fees would be if you should use their algorithms in some software that you plan to distribute or sell.

It depends on the application and cost. The ability to customize code trumps all else, particularly for statistical libraries, etc. But I'm not going to become an expert in optimization or code vectorization when there are free libraries that are readily adaptable (for example, numpy, scipy and many free libraries in R). Some combination is surely the sweet spot, but it's highly variable for your needs.