def _stats(self, a, b):
nA, nB = self._cdfa, self._cdfb
d = nB - nA
pA, pB = crv_helper._norm_pdf(a), crv_helper._norm_pdf(b)
mu = (pA - pB) / d # correction sign
mu2 = 1 + (a * pA - b * pB) / d - mu * mu
return mu, mu2, None, None
def _pdf(self, x, *args):
accum = np.zeros((len(x)))
for pos, prob in zip(self.discrete_d.xk, self.discrete_d.pk):
x_transl = (x - pos) / self.smooth_scale
accum = np.asarray([
acc + prob * _norm_pdf(x_) for acc, x_ in zip(accum, x_transl)
])
# accum /= len(self.discrete_d.xk)
return accum
def _pdf(self, x):
return _norm_pdf((x - self.mean) / self.sigma) / self.sigma / self.norm
def _pdf(self, x, a, b):
return crv_helper._norm_pdf(x) / self._delta
def _pdf(self, x):
return _norm_pdf((x - self.mean) / self.sigma) / self.sigma / self.norm
def ndtri_grad_hess(x):
f = ndtri(x)
phi = _norm_pdf(f)
grad = np.reciprocal(phi)
hess = grad ** 2 * f
return f, grad, hess
def ndtri_grad(x):
return np.reciprocal(_norm_pdf(ndtri(x)))
