core.feature Module

This module contains code for extracting features from data

exception skbeam.core.feature.PeakRejection

Bases: Exception

Custom exception class to indicate that the refine function rejected the candidate peak.

This uses the exception handling framework in a method akin to StopIteration to indicate that there will be no return value.

skbeam.core.feature.filter_n_largest(y, cands, N)

Filters the N largest candidate peaks

Return a maximum of N largest candidates. If N > len(cands) then all of the cands will be returned sorted, else the indices of the N largest peaks will be returned in descending order.

Parameters

yarray

Independent variable

candsarray

An array containing the indices of candidate peaks

Nint

The maximum number of peaks to return, sorted by size. Must be positive

Returns

candsarray

An array of the indices of up to the N largest candidates

skbeam.core.feature.filter_peak_height(y, cands, thresh, window=5)

Filter to remove candidate that are too small. This is implemented by looking at the relative height (max - min) of the peak in a window around the candidate peak.

Parameters

yarray

Independent variable

candsarray

An array containing the indices of candidate peaks

threshint

The minimum peak-to-peak size of the candidate peak to be accepted

windowint, optional

The size of the window around the peak to consider

Returns

candsarray

An array of the indices which pass the filter

skbeam.core.feature.peak_refinement(x, y, cands, window, refine_function, refine_args=None)

Refine candidate locations

Parameters

xarray

The independent variable, does not need to be evenly spaced.

yarray

The dependent variable. Must correspond 1:1 with the values in x

candsarray

Array of the indices in x (and y) for the candidate peaks.

refine_functionfunction

A function which takes a section of data with a peak in it and returns the location and height of the peak to sub-sample accuracy. Additional parameters can be passed through via the refine_args kwarg. The function signature must be:

center, height = refine_func(x, y, **kwargs)

This function may raise PeakRejection to indicate no suitable peak was found

windowint

How many samples to extract on either side of the candidate locations are passed to the refine function. The window will be truncated near the boundaries. The length of the data passed to the refine function will be (2 * window + 1).

refine_argsdict, optional

The passed to the refine_function

Returns

peak_locationsarray

The locations of the peaks

peak_heightsarray

The heights of the peaks

Examples

>>> x = np.arange(512)
>>> tt = np.zeros(512)
>>> tt += np.exp(-((x - 150.55)/10)**2)
>>> tt += np.exp(-((x - 450.75)/10)**2)
>>> cands = scipy.signal.argrelmax(tt)[0]

>>> print(peak_refinement(x, tt, cands, 10, refine_quadratic))
(array([ 150.62286432,  450.7909412 ]), array([ 0.96435832,  0.96491501]))
>>> print(peak_refinement(x, tt, cands, 10, refine_log_quadratic))
(array([ 150.55,  450.75]), array([ 1.,  1.]))

skbeam.core.feature.refine_log_quadratic(x, y, Rval_thresh=None)

Attempts to refine the peaks by fitting a quadratic to the log of the y-data. This is a linear approximation of fitting a Gaussian.

Parameters

xarray

Independent variable

yarray

Dependent variable

Rval_threshfloat, optional

Threshold for R^2 value of fit, If the computed R^2 is worse than this threshold PeakRejection will be raised

Returns

centerfloat

Refined estimate for center

heightfloat

Refined estimate for height

Raises

PeakRejection

Raised to indicate that no suitable peak was found in the interval

skbeam.core.feature.refine_quadratic(x, y, Rval_thresh=None)

Attempts to refine the peaks by fitting to a quadratic function.

Parameters

xarray

Independent variable

yarray

Dependent variable

Rval_threshfloat, optional

Threshold for R^2 value of fit, If the computed R^2 is worse than this threshold PeakRejection will be raised

Returns

centerfloat

Refined estimate for center

heightfloat

Refined estimate for height

Raises

PeakRejection

Raised to indicate that no suitable peak was found in the interval