Baseline fitting algorithm

[bas-rustenburg]

The currently implemented algorithm in python/bayesianitc/experiments does not perform well.

@pgrinaway experimented once with gaussian processes.

That said, we can also use origin files now to override the integrated heats. NITPIC also produces a .dat file with origin style formatting that ITC.py can read with the -q command line option.

[bas-rustenburg]

@pgrinaway just mentioned possibly using gptools.

[bas-rustenburg]

I've implemented a gaussian process fit in #22. For now, we still calculate the integrated heats, but we should add the baseline parameters to the model at a later date.

[bas-rustenburg]

Slight clarification: The choices for the gaussian process fitting parameters are arbitrary numbers that worked when I experimented with sklearn.

Here is where they are declared.

[jchodera]

I bet these can be selected by either cross-validation or (my preference) inference (using some hyperparameters) with some calibration datasets (e.g. long experiments without injections).