Dividing by zero when doing normalisation to prepare for Gridsearch

[young-x-skyee]

It seems that the problem of dividing by zero when doing normalisation for a constant signal in one second is not fixed in the main branch, but is already fixed in the language branch by using

def normalize(x: NDArray, eps: float = 1e-7) -> NDArray:
    """
    Remove the mean and divide by the Euclidean magnitude.
    """
    x -= np.mean(x, axis=-1, keepdims=True)
    x /= np.sqrt(np.sum(x**2, axis=-1, keepdims=True)) + eps
    return x

Maybe similar changes should be made to main as well?

[neukym]

Cai, I asked Tianyi to add this in as an issue because I wasn't sure whether the language branch was doing it in the same way as your fix - can you have a check and if it's already sorted in the main branch then we can close this.