Saving fit results to file (`LiveFitLogger`)

The LiveFitLogger callback exists to write the results of a LiveFit to file. These are designed to be human-readable files, rather than machine readable.

The files written by this callback contain:

The result of an lmfit.model.ModelResult.fit_report, which contains fit statistics such as r-squared, optimized values of fit parameters, and information about correlations if present.
Two blank lines
A csv table containing x, y, optionally y uncertainty, and modelled y

Example output file from this callback (click to expand)

[[Model]]
    Model(Gaussian  [amp * exp(-((x - x0) ** 2) / (2 * sigma**2)) + background])
[[Fit Statistics]]
    # fitting method   = leastsq
    # function evals   = 36
    # data points      = 98
    # variables        = 4
    chi-square         = 148.424352
    reduced chi-square = 1.57898247
    Akaike info crit   = 48.6805775
    Bayesian info crit = 59.0204474
    R-squared          = 0.95949614
[[Variables]]
    amp:         230.617045 +/- 10.4151289 (4.52%) (init = 287)
    sigma:       0.00969209 +/- 2.7953e-04 (2.88%) (init = 0.009036228)
    x0:          0.49905249 +/- 3.5203e-04 (0.07%) (init = 0.4999994)
    background:  0.59641742 +/- 0.13909858 (23.32%) (init = 0)
[[Correlations]] (unreported correlations are < 0.100)
    C(amp, sigma) = -0.6254
    C(sigma, x0)  = +0.1980
    C(amp, x0)    = -0.1360


x,y,y uncertainty,modelled y
0.29442974330116267,0.9362155048960608,1.1732576103594294,0.596417422655042
0.2987958123011627,1.7278474231222112,1.411659205995935,0.596417422655042
0.3013446053011628,2.2956249406851517,1.770613464899424,0.596417422655042
0.30558076430116277,3.574729064039327,2.1816903049886642,0.596417422655042
0.31062594030116286,1.0235264249142935,1.2440461004858383,0.596417422655042
...

Example configuration (click to expand)

def some_plan() -> Generator[Msg, None, None]:
    ... # Set up prefix, reducers, controllers etc. here

    @subs_decorator(
        [
            LiveFitLogger(
                lf, # LiveFit
                y=reducer.intensity.name,
                x=block.name,
                output_dir=Path(f"C:\\Instrument\\Var\\logs\\bluesky\\fitting"),
                postfix="bob" # Make sure to have different postfixes if using 
                    # more than 1 LiveFitLogger per run
                yerr=reducer.intensity_stddev.name, # Not required
            )
            ... # Other callbacks ie. live table/plot here - you can use multiple!
        ]
    )
    def _inner() -> Generator[Msg, None, None]:
        ... # Continue to plan

This will write data describing a fit to a file, into the path provided to the callback. If multiple fits are present simultaneously, use the postfix argument to append to the end of the filename to disambiguate different fits and to avoid overwriting fitting files.

If you provide a signal name for the yerr argument, then an extra column for y uncertainty will be displayed in the fitting file. Keep in mind that even if you provide yerr in LiveFitLogger, you will still also need to provide yerr in LiveFit if you want uncertainty/weight per point to influence the fit.

Saving fit results to file (LiveFitLogger)

Saving fit results to file (`LiveFitLogger`)