sf_kmeans.py 文件源码

def log_likelihood(self, data):
        nks = np.bincount(self.labels_, minlength=self.n_clusters)  # number of points in each cluster
        n, d = data.shape
        log_likelihood = 0
        covar_matrices = self.covariances(self.labels_, cluster_centers=self.cluster_centers_, data=data)
        covar_matrix_det_v = np.linalg.det(covar_matrices)
        self._inv_covar_matrices = self._matrix_inverses(covar_matrices)
        for k, nk in enumerate(nks):
            if self.verbose == 1:
                print('log_likelihood: covar_matrix_det = {}'.format(covar_matrix_det_v[k]))
            term_1 = nk * (np.log(float(nk)/n) - 0.5 * d * np.log(2*np.pi) - 0.5 * np.log(abs(covar_matrix_det_v[k])))
            cdist_result = cdist(data[self.labels_ == k], np.array([self.cluster_centers_[k]]), metric='mahalanobis', VI=self._inv_covar_matrices[k])
            cdist_no_nan = cdist_result[~np.isnan(cdist_result)]  #  to deal with nans returned by cdist
            term_2 = -0.5 * (np.sum(cdist_no_nan))
            k_sum = term_1 + term_2
            log_likelihood += k_sum
        if np.isnan(log_likelihood) or log_likelihood == float('inf'):
            raise Exception('ll is nan or inf')
        return log_likelihood