New peakshape

[Bisaloo]

This PR proposes a completely different algorithm to find the FWHM.

• The old algorithm estimated the FWHM by finding the closest values to the half maximum value. This meant that it could easily make mistakes when spectra had multiple peaks (see plots below).

  The new algorithm starts from the maximum value and then moves left (then right) until it finds the first value that falls below the half maximum value. This means it will never be fooled by multiple peaks

(left plot is the old algorithm and right plot is the new one)

• With the old algorithm, it was very had to determine when the FWHM overlapped with the wavelength range. This means that the right-hand side value (HWHM.r) was not actually at half maximum but was the lowest value to the right of the FWHM.

  With the new algorithm, it's trivial to detect such cases and peakshape() now accordingly returns NA when the FWHM range overlaps with the wavelength range.

(left plot is the old algorithm and right plot is the new one)

peakshape(sicalis, select = grepl("T$", colnames(sicalis)))

id	B3	H1	FWHM	HWHM.l	HWHM.r
ind1.T	18.38233	653	NA	145	NA
ind2.T	17.66600	681	NA	180	NA
ind3.T	14.53060	671	NA	165	NA
ind4.T	15.12063	660	NA	151	NA
ind5.T	17.49428	627	NA	123	NA
ind6.T	18.02172	632	NA	129	NA
ind7.T	16.76935	672	NA	164	NA

• The new algorithm is slower (spends 150% of the time spent before) than the old one but I may still be able to improve it a bit (fixed by ffeb760)

expr	min	lq	mean	median	uq	max	neval	cld
{ pavo::peakshape(sicalis, plot = FALSE) }	4.736414	4.813911	5.634903	4.873461	5.059997	186.11863	1000	a
{ peakshape(sicalis, plot = FALSE) }	4.697731	4.775466	5.324652	4.831968	4.966489	15.28568	1000	a

[thomased]

This looks awesome. Want to throw a few tests in if you get a chance? It's one of the last entirely-uncovered bits :)

[Bisaloo]

Sure! I'll probably have a go in a couple of days!