test_inference_plda.py 文件源码

def test_?_w(self):
        """ ?_w = [A][A^T]. See p. 533.

        DESCRIPTION: Since A is a free parameter, t will not necessarily recover
                      the original A. ?_w is what really describes the covariance
                      between cluster means (see p. 533), so that is what you want
                      to test - it is 'closer' to the data.
        """
        n_experiments = int(np.log10(1000000) / 2)
        n_list = [100 ** x for x in range(1, n_experiments + 1)]
        n_list = np.array(n_list).astype(float)
        n_dims = self.n_dims
        n_classes = 30 #self.n_classes

        ?_w_L1_errors = []
        for n in n_list:
            A, ?, model = self.experiment(int(n), n_dims, n_classes)

            ?_w = np.matmul(A, A.T)
            ?_w_model = np.matmul(model.A, model.A.T)

            L1_error = np.abs(?_w - ?_w_model).mean()
            abs_? = (np.abs(?_w).mean() + np.abs(?_w_model).mean()) * .5
            percent_error = L1_error / abs_? * 100
            print('Testing ?_w with {} samples: {} percent error'.format(n,
                  percent_error))
            ?_w_L1_errors.append(percent_error)

        Y = ?_w_L1_errors
        X = [x for x in range(len(?_w_L1_errors))]
        slope_of_error_vs_N = linregress(X, Y)[0]
        self.assertTrue(slope_of_error_vs_N < 0)