clf.py 文件源码

def compute_cmus_scov(self, data, labels):
        """ Compute class means and shared covariance
        (weighted within-class covariance) """

        self.label_map, class_sizes = np.unique(labels, return_counts=True)
        dim = data.shape[1]

        cmus = np.zeros(shape=(len(class_sizes), dim))
        acc = np.zeros(shape=(dim, dim))
        scov = np.zeros_like(acc)

        gl_mu = np.mean(data, axis=0).reshape(-1, 1)
        gl_cov = np.cov(data.T, bias=True)

        for i, k in enumerate(self.label_map):
            data_k = data[np.where(labels == k)[0], :]
            mu_k = np.mean(data_k, axis=0).reshape(-1, 1)
            cmus[i, :] = mu_k[:, 0]
            acc += (gl_mu - mu_k).dot((gl_mu - mu_k).T) * data_k.shape[0]

        acc /= data.shape[0]
        scov = gl_cov - acc

        return cmus, scov, class_sizes