TODO

* Put the plots on the same axis (raw)
* Plot normalized traces
* Find a better way to convert to long format that does not depend on region label
* Add a different baseline correction option for when isobestic point of the sensor is different from neurophotometric wavelength
* Biexponential decay (`biexponential decay`) fit in detrending function (demodulate) needs to work better, such that it works about as well as airPLS
* How to install `kaleido`
* Figure out why my iPython notebooks render white blocks instead of some text when using VStudio


* Rebuild batch processing for Rudi's workflow


from scipy.optimize import curve_fit

########
* Why is `p0_hypotheses` of `demodulate_with_biexponential_decay` hard-coded? On what basis are the initial values selected? Is there any reason not to randomly generate them each time? (setting a seed of course)

* airPLS and biexponential decay are methods of detrending, which corresponds to the "bleaching correction" step, right?

* What is the theoretical (or practical) reason for choosing preferring biexponential decay to airPLS and?

* If one "works" and the other doesn't, does that indicate an issue with the parameters, the data, or just that some data works better with one than the other?

* Is it kosher to use different parameter values for the two methods?

* Is the expectation that both methods give similar results? How important is that?

* Are `lambda`, `porder`, `itermax` the main parameters for artifact removal? Which of these are you comfortable playing with in order to get biexponential decay to work?