biospecnorway / biospectools Goto Github PK

View Code? Open in Web Editor NEW

3.0 3.0 3.0 2.19 MB

Tools for working with spectral data https://biospectools.readthedocs.io/en/latest/

Python 98.56% MATLAB 1.44%

biospectools's People

Contributors

Stargazers

Watchers

Forkers

victor-ne borondics freesiemens

biospectools's Issues

Fix norm_y_weights issue with new sklearn version

Introduce `interferents` and `analytes` parameters to EMSC models

Interferents is a renamed constituents paramter
Analytes are supposed to be as interferents, but eventually to be not removed from the spectrum
Support of old interface until 0.6.0

Rename parameter `internals` in EMSC models to `details`

Rename and support old kwargs until version 0.6.0

EMSC with multiregions does not construct polynomials in the good mid point

Problem

@JohanneSolheim spotted a problem that when EMSC is used with several non-connected regions of spectra, then the established polynomial components look to be off center. In the image below you can see that the red line (linear polynomial) is shifted more towards left region.

Explanation

Indeed, current implementation defines the middle point of spectrum as a mean of all wavenumbers (np.mean(wavenumbers)), what essentially means that the middle point lies in the center of mass of wavenumbers. This approach seems to be invalid when used with several non-connected spectral regions, because the middle point will depend on the number of points in each region. The edge case is when one region contains 1000 wavenumbers and another region contains only 1 wavenumber. In that case the mid-point will be placed in the first region.

The suggested improvement is to calculate the middle point as the average of min and max wavenumber ((wns.min() + wns.max()) / 2). In that case mid point will not depend on number of points in regions.

Affected routines

This improvement may change results of correction of spectra with glued multi-regions, for example lipid peaks around 3000 cm^-1 and 1700 cm^-1. For a single region it should work the same as previously.

Lines 72 to 113 in 2e2c3e5

 class EMSC: 

 """Extended multiplicative signal correction (EMSC) [1]_. 

  Parameters 

  ---------- 

  reference : `(K_channels,) ndarray` 

  Reference spectrum. 

  wavenumbers : `(K_channels,) ndarray`, optional 

  Wavenumbers must be passed if given polynomial order 

  is greater than zero. 

  poly_order : `int`, optional (default 2) 

  Order of polynomial to be used in regression model. If None 

  then polynomial will be not used. 

  weights : `(K_channels,) ndarray`, optional 

  Weights for spectra. 

  constituents : `(N_constituents, K_channels) np.ndarray`, optional 

  Chemical constituents for regression model [2]_. Can be used to add 

  orthogonal vectors. 

  scale : `bool`, default True 

  If True then spectra will be scaled to reference spectrum. 

  rebuild_model : `bool`, default True 

  If True, then model will be built each time transform is called, 

  this allows to dynamically change parameters of EMSC class. 

  Otherwise model will be built once (for speed). 

  Other Parameters 

  ---------------- 

  _model : `(K_channels, 1 + N_constituents + (poly_order + 1) ndarray` 

  Matrix that is used to solve least squares. First column is a 

  reference spectrum followed by constituents and polynomial columns. 

  _norm_wns : `(K_channels,) ndarray` 

  Normalized wavenumbers to -1, 1 range 

  References 

  ---------- 

  .. [1] A. Kohler et al. *EMSC: Extended multiplicative 

  signal correction as a tool for separation and 

  characterization of physical and chemical 

  information in fourier transform infrared 

  microscopy images of cryo-sections of 

  beef loin.* Applied spectroscopy, 59(6):707–716, 2005. 

  """

Array shape mismatch out of DSAE

Hi,
I'm running into an error with the DSAE. When I try to correct spectra, I get an output which is longer by 1 than my raw spectra, so it can't be reshaped :

Python version : 3.8.12
Biospectools version : 0.3.1

	class EMSC:
	"""Extended multiplicative signal correction (EMSC) [1]_.

	Parameters
	----------
	reference : `(K_channels,) ndarray`
	Reference spectrum.
	wavenumbers : `(K_channels,) ndarray`, optional
	Wavenumbers must be passed if given polynomial order
	is greater than zero.
	poly_order : `int`, optional (default 2)
	Order of polynomial to be used in regression model. If None
	then polynomial will be not used.
	weights : `(K_channels,) ndarray`, optional
	Weights for spectra.
	constituents : `(N_constituents, K_channels) np.ndarray`, optional
	Chemical constituents for regression model [2]_. Can be used to add
	orthogonal vectors.
	scale : `bool`, default True
	If True then spectra will be scaled to reference spectrum.
	rebuild_model : `bool`, default True
	If True, then model will be built each time transform is called,
	this allows to dynamically change parameters of EMSC class.
	Otherwise model will be built once (for speed).

	Other Parameters
	----------------
	_model : `(K_channels, 1 + N_constituents + (poly_order + 1) ndarray`
	Matrix that is used to solve least squares. First column is a
	reference spectrum followed by constituents and polynomial columns.
	_norm_wns : `(K_channels,) ndarray`
	Normalized wavenumbers to -1, 1 range

	References
	----------
	.. [1] A. Kohler et al. *EMSC: Extended multiplicative
	signal correction as a tool for separation and
	characterization of physical and chemical
	information in fourier transform infrared
	microscopy images of cryo-sections of
	beef loin.* Applied spectroscopy, 59(6):707–716, 2005.
	"""