pi, degrees, radians

[teuler]

Hi, maybe I overlooked it, but it would be nice to have pi, and for code clarity also degrees() and radians() in ulab. The idea is that ideally, one would import ulab for all math instead of additionally having to import math. Thanks

[v923z]

I haven't included these, because, as you point out, they are either in the standard math module, or are trivial. On the other hand, I see your point.

[v923z]

@teuler https://github.com/v923z/micropython-ulab/pull/163 implements the requested changes. The PR adds approx. 600 bytes to the firmware. I am a bit uncertain as to whether it is worth it for things that are already available.

The two constants add 96 bytes (a bit odd), and the two functions another 500. The thing is, degrees is really nothing more than

a = ulab.array([1, 2, 3, 4])
b = (180.0/ulab.pi) * a

In fact, I haven't measured, but I am positive that the conversion in the last line is actually faster than degrees, because degrees contains function calls in its for loop, while b = (180.0/ulab.pi) * a is just a binary operation, with straight multiplications, and without any function calls in the loop.

@jepler Before merging, I would like to hear your opinion.

[teuler]

@v923z Thanks for the quick response. I guess it comes down to the question how often 'ulab' can fully replace importing 'math'; then again, 'math' is already in the firmware. I tend to say that the added memory load is not worth the convenience of not having to import 'math'.

[teuler]

Background of my question: I have code that runs both in MicroPython and normal Python (w/ numpy). Using import ulab as np or import numpy as np allows more elegant code :)

[v923z]

Background of my question: I have code that runs both in MicroPython and normal Python (w/ 'numpy'). Using 'import ulab as np' or 'import numpy as np' allows more elegant code :)

That could be, but ulab is split into sub-modules, so, e.g., if you want to evaluate a polynomial, then you have to

from ulab import poly

poly.polyval(p, x)

which under numpy is simply

import numpy as np

np.polyval(p, x)

We defined sub-modules, so that partial firmware can be compiled for smaller systems: https://micropython-ulab.readthedocs.io/en/latest/ulab.html#customising-ulab If you are willing to forgo the modularity (i.e., want to compile all modules, or you don't care, if your code throws an error at runtime), I could probably do this as in numpy, and you could then set a pre-processor switch for numpy-like or ulab-like syntax. The problem is not really the code, but the documentation. It could become a tad confusing.

[v923z]

This, and a lot of other features have been added in version 1.0.0. Closing the issue now.