def _kld(self, data, weights=None, normalize=False):
assert((data >= 0).all() and (data <= self.n).all())
alphas,betas = self._get_abs(data, weights)
result = -dist_fit.log_beta_binomial( data, self.n, alphas, betas )
if normalize:
# evs = self.n*alphas/(alphas + betas)
raise NotImplementedError('not sure how to normalize likelihood for discrete distribution')
return result
def kld(self, succ, att):
if att == 0: return 0.
if succ > att: att = succ
return -dist_fit.log_beta_binomial(succ, att, *self.ab)
def chi_sq(self, succ, att):
# if att == 0: return 0.
norm = dist_fit.log_beta_binomial(self.ev(att), att, *self.ab)
return 2. * (self.kld(succ, att) + norm)