def cdf_grid(self):
if self._cdf_grid is None:
if self._FINITE:
pad = (1 + _NUM_PAD)
args = [-pad, pad]
else:
args = [-6, 6]
num = np.diff(args)[0] * _INTERP_NUM_PER_STD
args = args + [int(num), ]
xc = np.linspace(*args)
if self._CDF_INTERP:
yy = self.evaluate(xc)
csum = utils.cumtrapz(yy, xc, prepend=False)
else:
csum = self.cdf(xc)
norm = csum[-1]
if not np.isclose(norm, 1.0, rtol=1e-5):
err = "Failed to reach unitarity in CDF grid norm: {:.4e}!".format(norm)
raise ValueError(err)
# csum = csum / norm
# xc = np.concatenate([[args[0]], [args[0]], xc, [args[1]], [args[1]]], axis=0)
# csum = np.concatenate([[0.0 - _NUM_PAD], [0.0], csum, [1.0], [1.0+_NUM_PAD]], axis=0)
self._cdf_grid = [xc, csum]
return self._cdf_grid
def cdf(self, pnts, params=None, reflect=None):
"""Cumulative Distribution Function based on KDE smoothed data.
Arguments
---------
pnts : ([D,]N,) array_like of scalar
Target evaluation points
Returns
-------
cdf : (N,) ndarray of scalar
CDF Values at the target points
"""
if params is not None:
raise NotImplementedError(
"`params` is not yet implemented for CDF!")
if reflect is not None:
raise NotImplementedError(
"`reflect` is not yet implemented for CDF!")
if self._cdf_func is None:
points = self.points
# Calculate PDF at grid locations
pdf = self.pdf(points, grid=True)[1]
# Convert to CDF using trapezoid rule
cdf = utils.cumtrapz(pdf, points)
# Normalize to the maximum value
cdf /= cdf.max()
ndim = np.ndim(cdf)
# points = np.atleast_2d(points)
if ndim == 1 and np.ndim(points) == 1:
points = np.atleast_2d(points)
self._cdf_grid = (points, cdf)
self._cdf_func = sp.interpolate.RegularGridInterpolator(
*self._cdf_grid, bounds_error=False, fill_value=None)
# `scipy.interplate.RegularGridInterpolator` expects shape (N,D,) -- so transpose
pnts = np.asarray(pnts).T
cdf = self._cdf_func(pnts)
return cdf