distributions.py 文件源码

def max_likelihood(self,data,weights=None,stats=None):
        if stats is not None:
            n, tot = stats
        elif weights is None:
            n, tot = super(NegativeBinomialIntegerR,self)._get_statistics(data)
        else:
            n, tot = super(NegativeBinomialIntegerR,self)._get_weighted_statistics(data,weights)

        if n > 1:
            rs = self.r_support
            ps = self._max_likelihood_ps(n,tot,rs)

            # TODO TODO this isn't right for weighted data: do weighted sums
            if isinstance(data,np.ndarray):
                likelihoods = np.array([self.log_likelihood(data,r=r,p=p).sum()
                                            for r,p in zip(rs,ps)])
            else:
                likelihoods = np.array([sum(self.log_likelihood(d,r=r,p=p).sum()
                                            for d in data) for r,p in zip(rs,ps)])

            argmax = likelihoods.argmax()
            self.r = self.r_support[argmax]
            self.p = ps[argmax]
        return self