Global params can be Numpy arrays

[aleksandra-kim]

Original report by Chris Mutel (Bitbucket: cmutel, GitHub: cmutel).

---

Don't only validate against Number

[aleksandra-kim]

Original comment by Pascal Lesage (Bitbucket: MPa, ).

---

This assumes that the array is stored in the amount field. This makes it impossible to store both the deterministic value and a sampled array. Perhaps the Numpy array needs to be stored in a new field, but then we'd have to make sure that these sampled arrays can be passed to the Interpreter.

[BenPortner]

@cmutel The original issue is no longer accessible. Can you reproduce it from the top of your head?

[cmutel]

Sure; we currently assume that a variable can only have a single numeric value, but for computational efficiency it would be nice if they could have an array of values so that we would only need to evaluate the string formulas once to get e.g. 1000 results. But this is a bit tricky, you need a convention to make sure the numpy arrays broadcast correctly, and write the formulas so that they can accept numpy arrays with leading to everything being NaN if there is an error with one value in the whole array (e.g. if you would normalize by the sum or average).