I can't find online_semg_posture_adaptation.online.covariance

[MarcTasca]

Dear authors,

I am writing my thesis at PoliTo. First of all, thank you a lot for making this repository public. I would love to know the implementation of covariance, which should be in online_semg_posture_adaptation/online, but I can't find it. In particular, in online_semg_posture_adaptation/online/pca.py there is the function oja_sga_session( ... ), in here it is called, at line 158 the function cov.update_cov( ... ): mean, ncov, _ = cov.update_cov(x[:, idx_sample], mean, ncov, idx_sample + 1)

This should be imported in the file pca.py at line 27: from online_semg_posture_adaptation.online import covariance as cov.

I can't find it. Thank you a lot and best regards :)

[MarcTasca]

This is my implementation.

My covariance.py:

def update_cov(
    x: np.ndarray[np.float32],
    mean: np.ndarray[np.float32],
    ncov: np.ndarray[np.float32],
    count: int
) -> Tuple[
        np.ndarray[np.float32], 
        np.ndarray[np.float32], 
        int]:
    """
    Update the mean and covariance online.

    Args:
    x: np.ndarray[np.float32]: the new sample
    mean: np.ndarray[np.float32]: the current mean
    ncov: np.ndarray[np.float32]: the current covariance
    count: int: the current number of samples

    Returns:
    np.ndarray[np.float32]: the updated mean
    np.ndarray[np.float32]: the updated covariance
    int: the updated number of samples
    """
    # update mean and cov online
    mean = mean + (x - mean) / count
    ncov = ncov + np.outer(x - mean, x - mean)

    # return them
    return mean, ncov, count

In online_semg_posture_adaptation.online.pca.oja_sga_session I added:

scale = np.sqrt(np.diag(ncov) / (idx_sample + 1))
Scale[scale == 0] = 1

If I don't do so, all W result to be filled with nan values.

Is that similar to your implementation?