arviz_stats.base.array_stats.kde

array_stats.kde(ary, axis=-1, circular=False, grid_len=512, **kwargs)

Compute KDE on array-like inputs.

Parameters:

aryarray_like

axisint, sequence of int or None, default -1

circularbool, default False

grid_lenint, default 512

**kwargs

Additional keyword arguments passed to the underlying KDE implementation. Depending on whether circular is True or False, supported arguments include:

• bwint, float, or str, optional

  The bandwidth or the method to estimate it. Options include “scott”, “silverman”, “isj”, “experimental”, or “taylor” (if circular). Defaults to “experimental” (or “taylor” if circular).

• bw_fctfloat, optional

  A multiplier for bw to tune smoothness manually. Defaults to 1.

• adaptivebool, optional

  If True, uses an adaptive bandwidth. Not compatible with circular KDE. Defaults to False.

• extendbool, optional

  If True, extends the observed range for linear KDE. Defaults to False.

• extend_fctfloat, optional

  Number of standard deviations to widen bounds if extend is True. Defaults to 0.5.

• bound_correctionbool, optional

  Whether to perform boundary correction on linear bounds. Defaults to True.

• custom_limslist or tuple, optional

  Custom bounds for the range of ary. Defaults to None.

• cumulativebool, optional

  If True, returns the CDF instead of the PDF. Defaults to False.

Returns:

grid, pdf, bwarray_like

grid and pdf will have the same shape: the same as ary minus the dimensions in axis plus an extra dimension of length grid_len. Same for bw except it will not have the extra dimension.