Sweep Documentation¶
Sweeping in measurement software is often a very rigid afair. Typical parameter sweeps take the form
for this in range(10):
for that in np.linspace(30.3, 100.6, 27):
set_this(this)
set_that(that)
measure_everything()
A few questions arise at this point:
- Is there any reason to use a rectangular grid of points?
- Am I wasting time if features aren’t uniformly distributed over this grid?
- What if our range wasn’t sufficient to capture the desired data?
- What if we didn’t get good enough statistics?
To tackle (1), there are some clear reasons to measure on a rectilinear grid. First of all, it is extremely convenient. Also, if you expect that regions of interest (ROI) are distributed evenly across your measurement domain then this seems like a reasonable choice. Finally there are simple aesthetic considerations: image plots look much better when they are fill a rectangular domain rather than leaving swaths of NaNs strewn the periphery of your image.
Point (2) is really an extension of (1): if you are looking at data that follows sin(x)*sin(y) then the information density is practically constant across the domain. If you are looking at a crooked phase transition in a binary system, then the vast majority of your points will be wasted on regions with very low information content. Take the following phase diagrams for an MRAM cell’s switching probability.