Structure function implementation is really complicated
The current structure function implementation looks really complicated and seems to be using a slow for-loop.
Isn't the structure function just the corellation function in the end?
from the meeting:
should translate to:
\frac{structurefunction_{signal}(r)}{2 \sigma ^ 2} = 1 - autocorellationfunction_{signal}(r)
There is np.correlate
which does this already and very efficiently.
Efficiency came up for Sary's analysis where she has to do a moving window of structure parameter calculation. She ported Andy's R/C-code to Python (slow for-loops all over the place) and applied to a million data points, even on beefy it's unbearably slow. She'll now test parmesan's implementation and see if it's faster (and the same