171afc33723ad7f0a061131ad85041cdf8f305e5,nara_wpe/wpe.py,,get_filter_matrix_conjugate_v3,#Any#Any#Any#Any#,120

Before Change

        Y, inverse_power, K, delay
    )

    inv = np.linalg.inv(correlation_matrix)
    inv = np.kron(np.eye(D), inv)

    correlation_vector = np.reshape(correlation_vector, (D * D * K, 1))

    selector = np.transpose(np.reshape(
        np.arange(D * D * K), (-1, K, D)
    ), (1, 0, 2)).flatten()

    inv = inv[:, selector]
    inv = inv[selector, :]

    stacked_filter_conj = inv @ correlation_vector
    filter_matrix_conj = np.transpose(
        np.reshape(stacked_filter_conj, (K, D, D)), (0, 2, 1))
    return filter_matrix_conj

After Change

    // This should still be faster and more stable than np.linalg.inv().
    // print(np.linalg.cond(correlation_matrix))
    stacked_filter_conj = np.reshape(
        np.linalg.solve(
            correlation_matrix[None, :, :],
            np.reshape(correlation_vector, (D, D * K, 1))
        ),
        (D * D * K, 1)
    )
    stacked_filter_conj = stacked_filter_conj[selector, :]

Instances