here's a web app I made that generates code for efficiently approximating mathematical functions. This is useful when performance matters more than perfect accuracy, for example in embedded systems.
The app uses Chebyshev expansions, which despite their theoretical depth result in suprisingly compact and readable code in practice. This code is generated for you and using it does not require any knowledge of the underlying theory.
Source code and more info: https://github.com/stuffmatic/chebyshev-calculator
Here's a rather wonderful original document from the BBC Research Department (I had no idea that was a thing) back in 1969 going over just what makes them so great (https://downloads.bbc.co.uk/rd/pubs/reports/1969-10.pdf).
If all you've come across are Taylor approximations, these things can seem a little like magic at first.
Doesn't handle divide by zero very well though i.e. f(x)=1/x. Should probably consider that as undefined rather than a bad expression.
alternatively subset of the numerical analysi wikipedia page or these algorithms in the book is a good seed filter [2]
ideally covers some linalg like gaussian elim, power method, newton's root finding dynamic, and issues like approximating by discretization, recursion to reduce solve iterations, and things like convergence due to numerical instability due to IEEE 754 fp16 limits etc
1. https://en.wikipedia.org/wiki/Numerical_analysis
2. https://chatgpt.com/share/67002e8c-c5c4-8008-83f9-a6ebc603ad...
Faster Math Functions:
https://basesandframes.files.wordpress.com/2016/05/fast-math...
https://basesandframes.files.wordpress.com/2016/05/fast-math...
Even faster math functions GDC 2020:
https://gdcvault.com/play/1026734/Math-in-Game-Development-S...
To work around it, I handled the x near zero case by just forcing to 1.0.
if(Math.abs(x) > 1e-8 ){ Math.sin(x)/x } else { 1.0 }
That wouldn't exactly be a bug. The code is undoubtedly calculating the Chebyshev coefficients by evaluating the function on something like x_j = (xmin) + (xmax - xmin)/2(1 + cos(pi[0..j-1]/(j-1)). If one of those grid points happens to be exactly 0, it will try to evaluate Math.sin(0)/0, giving the NaN.
Another workaround is to have a slightly asymmetric range, such as [-3,+3.0000001]
I'd like to generate a Chebyshev approximation for a set of X, Y sensor values. Any hints on how to modify your code to do that?
If you have a few spare CPU cycles, a hybrid approximation could start with a sparse lookup table of values as the initial guess for a few rounds of a numerical approximation technique. Or you just store the first few coefficients of a polynomial approximation (as in the OP's work).
Training a neural net to calculate sines is like the math equivalent of using an LLM to reverse a string. Sure, you *can*, but the idea only makes sense if you don't understand how fundamentally solvable the problem is with a more direct approach.
It's always worth looking if mathematicians already have a solution to a problem before reaching for AI/ML techniques. Unfortunately, a lot of effort is probably being spent these days programming some kind of AI/ML to solve problems that have a known, efficient, maybe even proven optimal solution that developers just don't know about.
Neural nets can be useful when you have samples of a function but no idea how to approximate it, but that's not the case here.
The main strength of a neural network comes into play when there's a lot of different inputs, not just a handful. For the simpler cases like sin(x) we have other tools like the one posted here.
[1]: https://www.youtube.com/watch?v=FBpPjjhJGhk But what is a neural network REALLY?
Note: results. The software itself is a bit of a pain to use.
One’s first go to method should be Chebyshev. Neural nets used as a last resort.
Any chance you can add a rational function version?
import numpy as np
p = np.polynomial.Chebyshev.interpolate(f, degree, domain=(xmin, xmax))
# insert your code to print out some C code
Also strongly recommend some basic familiarity with the theory. Approximating `Math.abs(x)` to even a few digits of uniform accuracy on any interval containing 0 requires tens if not hundreds of thousands of coefficients. # insert your code to print out some C code
mean?I could do:
print('double coef[] = {')
for c in p.coef:
print(f' {c:0.16g},')
print('};')
and copy-paste it wherever I need.For many purposes it's much better to represent a polynomial as a combination of Chebyshev polynomials: 1, x, 2x^2-1, 4x^3-3x, etc.
(Supposing you are primarily interested in values of x between -1 and +1. For other finite intervals, use Chebyshev polynomials but rescale x. If x can get unboundedly large, consider whether polynomials are really the best representation for the functions you're approximating.)
Handwavy account of why: Those powers of x are uncomfortably similar to one another; if you look at, say, x^4 and x^6, they are both rather close to 0 for smallish x and shoot up towards 1 once x gets close to +-1. So if you have a function whose behaviour is substantially unlike these and represent it as a polynomial, you're going to be relying on having those powers largely "cancel one another out", which means e.g. that when you evaluate your function you'll often be representing a smallish number as a combination of much larger numbers, which means you lose a lot of precision.
For instance, the function cos(10x) has 7 extrema between x=-1 and x=+1, so you should expect it to be reasonably well approximated by a polynomial of degree not too much bigger than 8. In fact you get a kinda-tolerable approximation with degree 12, and the coefficients of the best-fitting polynomial when represented as a combination of Chebyshev polynomials are all between -1 and +1. So far, so good.
If we represent the same function as a combination of powers, the odd-numbered coefficients are zero (as are those when we use the Chebyshev basis; in both cases this is because our function is an even function -- i.e., f(-x) = f(x)), but the even-numbered ones are now approximately 0.975, -4.733, 370.605, -1085.399, 1494.822, -994.178, 259.653. So we're representing this function that takes values between -1 and +1 as a sum of terms that take values in the thousands!
(Note: this isn't actually exactly the best-fitting function; I took a cheaty shortcut to produce something similar to not quite equal to the minimax fit. Also, I make a lot of mistakes and maybe there are some above. But the overall shape of the thing is definitely as I have described.)
Since our coefficients will be stored only to some finite precision, this means that when we compute the result we will be losing several digits of accuracy.
(In this particular case that's fairly meaningless, because when I said "kinda-tolerable" I meant it; the worst-case errors are on the order of 0.03, so losing a few places of accuracy in the calculation won't make much difference. But if we use higher-degree polynomials for better accuracy and work in single-precision floating point -- as e.g. we might do if we were doing our calculations on a GPU for speed -- then the difference may really bite us.)
It also means that if we want a lower-degree approximation we'll have to compute it from scratch, whereas if we take a high-degree Chebyshev-polynomial approximation and just truncate it by throwing out the highest-order terms it usually produces a result very similar to doing the lower-degree calculation from scratch.