sigma_clip

astropy.stats.sigma_clip(data, sigma=3, sigma_lower=None, sigma_upper=None, maxiters=5, cenfunc='median', stdfunc='std', axis=None, masked=True, return_bounds=False, copy=True, grow=False)

Perform sigma-clipping on the provided data.

The data will be iterated over, each time rejecting values that are less or more than a specified number of standard deviations from a center value.

Clipped (rejected) pixels are those where:

data < center - (sigma_lower * std)
data > center + (sigma_upper * std)

where:

center = cenfunc(data [, axis=])
std = stdfunc(data [, axis=])

Invalid data values (i.e., NaN or inf) are automatically clipped.

For an object-oriented interface to sigma clipping, see SigmaClip.

Note

scipy.stats.sigmaclip provides a subset of the functionality in this class. Also, its input data cannot be a masked array and it does not handle data that contains invalid values (i.e., NaN or inf). Also note that it uses the mean as the centering function. The equivalent settings to scipy.stats.sigmaclip are:

sigma_clip(sigma=4., cenfunc='mean', maxiters=None, axis=None,
...        masked=False, return_bounds=True)

Parameters:

datanumpy:array_like or MaskedArray

The data to be sigma clipped.

sigmapython:float, optional

The number of standard deviations to use for both the lower and upper clipping limit. These limits are overridden by sigma_lower and sigma_upper, if input. The default is 3.

sigma_lowerpython:float or python:None, optional

The number of standard deviations to use as the lower bound for the clipping limit. If None then the value of sigma is used. The default is None.

sigma_upperpython:float or python:None, optional

The number of standard deviations to use as the upper bound for the clipping limit. If None then the value of sigma is used. The default is None.

maxiterspython:int or python:None, optional

The maximum number of sigma-clipping iterations to perform or None to clip until convergence is achieved (i.e., iterate until the last iteration clips nothing). If convergence is achieved prior to maxiters iterations, the clipping iterations will stop. The default is 5.

cenfunc{‘median’, ‘mean’} or python:callable(), optional

The statistic or callable function/object used to compute the center value for the clipping. If using a callable function/object and the axis keyword is used, then it must be able to ignore NaNs (e.g., numpy.nanmean) and it must have an axis keyword to return an array with axis dimension(s) removed. The default is 'median'.

stdfunc{‘std’, ‘mad_std’} or python:callable(), optional

The statistic or callable function/object used to compute the standard deviation about the center value. If using a callable function/object and the axis keyword is used, then it must be able to ignore NaNs (e.g., numpy.nanstd) and it must have an axis keyword to return an array with axis dimension(s) removed. The default is 'std'.

axispython:None or python:int or python:tuple of python:int, optional

The axis or axes along which to sigma clip the data. If None, then the flattened data will be used. axis is passed to the cenfunc and stdfunc. The default is None.

maskedbool, optional

If True, then a MaskedArray is returned, where the mask is True for clipped values. If False, then a ndarray and the minimum and maximum clipping thresholds are returned. The default is True.

return_boundsbool, optional

If True, then the minimum and maximum clipping bounds are also returned.

copybool, optional

If True, then the data array will be copied. If False and masked=True, then the returned masked array data will contain the same array as the input data (if data is a ndarray or MaskedArray). If False and masked=False, the input data is modified in-place. The default is True.

growpython:float or False, optional

Radius within which to mask the neighbouring pixels of those that fall outwith the clipping limits (only applied along axis, if specified). As an example, for a 2D image a value of 1 will mask the nearest pixels in a cross pattern around each deviant pixel, while 1.5 will also reject the nearest diagonal neighbours and so on.

Returns:

resultnumpy:array_like

If masked=True, then a MaskedArray is returned, where the mask is True for clipped values and where the input mask was True.

If masked=False, then a ndarray is returned.

If return_bounds=True, then in addition to the masked array or array above, the minimum and maximum clipping bounds are returned.

If masked=False and axis=None, then the output array is a flattened 1D ndarray where the clipped values have been removed. If return_bounds=True then the returned minimum and maximum thresholds are scalars.

If masked=False and axis is specified, then the output ndarray will have the same shape as the input data and contain np.nan where values were clipped. If the input data was a masked array, then the output ndarray will also contain np.nan where the input mask was True. If return_bounds=True then the returned minimum and maximum clipping thresholds will be be ndarrays.

See also

SigmaClip, sigma_clipped_stats

Notes

The best performance will typically be obtained by setting cenfunc and stdfunc to one of the built-in functions specified as as string. If one of the options is set to a string while the other has a custom callable, you may in some cases see better performance if you have the bottleneck package installed.

Examples

This example uses a data array of random variates from a Gaussian distribution. We clip all points that are more than 2 sample standard deviations from the median. The result is a masked array, where the mask is True for clipped data:

>>> from astropy.stats import sigma_clip
>>> from numpy.random import randn
>>> randvar = randn(10000)
>>> filtered_data = sigma_clip(randvar, sigma=2, maxiters=5)

This example clips all points that are more than 3 sigma relative to the sample mean, clips until convergence, returns an unmasked ndarray, and does not copy the data:

>>> from astropy.stats import sigma_clip
>>> from numpy.random import randn
>>> from numpy import mean
>>> randvar = randn(10000)
>>> filtered_data = sigma_clip(randvar, sigma=3, maxiters=None,
...                            cenfunc=mean, masked=False, copy=False)

This example sigma clips along one axis:

>>> from astropy.stats import sigma_clip
>>> from numpy.random import normal
>>> from numpy import arange, diag, ones
>>> data = arange(5) + normal(0., 0.05, (5, 5)) + diag(ones(5))
>>> filtered_data = sigma_clip(data, sigma=2.3, axis=0)

Note that along the other axis, no points would be clipped, as the standard deviation is higher.