[WIP] Display/return quantities with units

[mstimberg]

Internally, all values are unitless, but now values that are returned to the user are displayed with units. This affects:

• the returned parameters and errors from Fitter.fit and Fitter.refine
• the values stored in Fitter.best_params and Fitter.best_error
• the values returned by Fitter.results (except when using dataframe, since it doesn't support units)
• The parameters/error displayed in the callback functions

This of course might break some existing scripts (or give weird results if they multiply things with units again). You can create your Fitter with use_units=False to switch back to the previous behaviour. I think that given that the package is still "beta" and not widely used (I guess), it is still ok to break things this way. I prefer doing this now and having use_units=True as the default instead of having to tell users to add an option to get the nicer behaviour.

There's one thing that still needs to be fixed: the callback gets the normalized error value, but the callback does not know about this and will display it wrong. I'm not quite sure what's the best solution here. Would it be more natural for the callback to show the error normalized without units (e.g. if the actual error is 2mV and your normalization is 1mV, it would be displayed as 2), or should it revert the normalization, i.e. always show the same error independent of normalization? @romainbrette any thoughts?

Closes #18, closes #19

[romainbrette]

I would think that normalized error is better. The most important point, in any case, is that it is consistent between the population-based method and the gradient method.

[mstimberg]

This is what it looks like now:

with use_units=True (default)

no normalization (normalization=1):

... (error: 37.62624665 mV^2)

normalization=1*mV:

... (error: 37.62624665)

normalization=1e-3:

(error: 37.62624665 V^2)

 

with use_units=False

no normalization (normalization=1):

... (error: 3.762624665e-05)

normalization=1*mV or normalization=1e-3 (no difference) :

... (error: 37.62624665)

Using a normalization without units and calc_units=True certainly looks odd (3rd example above), but in a way it makes sense? We could also simply not allow a normalization without units if you use units in general. What do you think?

Either way all this is consistent between fit and refine.

[romainbrette]

I think it's all fine, it's what I would expect.