graspologic/embed/svd.py [19:57]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
def _compute_likelihood(arr: np.ndarray) -> np.ndarray:
    """
    Computes the log likelihoods based on normal distribution given
    a 1d-array of sorted values. If the input has no variance,
    the likelihood will be nan.
    """
    n_elements = len(arr)
    likelihoods = np.zeros(n_elements)

    for idx in range(1, n_elements + 1):
        # split into two samples
        s1 = arr[:idx]
        s2 = arr[idx:]

        # deal with when input only has 2 elements
        if (s1.size == 1) & (s2.size == 1):
            likelihoods[idx - 1] = -np.inf
            continue

        # compute means
        mu1 = np.mean(s1)
        if s2.size != 0:
            mu2 = np.mean(s2)
        else:
            # Prevent numpy warning for taking mean of empty array
            mu2 = -np.inf

        # compute pooled variance
        variance = ((np.sum((s1 - mu1) ** 2) + np.sum((s2 - mu2) ** 2))) / (
            n_elements - 1 - (idx < n_elements)
        )
        std = np.sqrt(variance)

        # compute log likelihoods
        likelihoods[idx - 1] = np.sum(norm.logpdf(s1, loc=mu1, scale=std)) + np.sum(
            norm.logpdf(s2, loc=mu2, scale=std)
        )

    return likelihoods
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



graspologic/pipeline/embed/_elbow.py [13:51]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
def _compute_likelihood(arr: np.ndarray) -> np.ndarray:
    """
    Computes the log likelihoods based on normal distribution given
    a 1d-array of sorted values. If the input has no variance,
    the likelihood will be nan.
    """
    n_elements = len(arr)
    likelihoods = np.zeros(n_elements)

    for idx in range(1, n_elements + 1):
        # split into two samples
        s1 = arr[:idx]
        s2 = arr[idx:]

        # deal with when input only has 2 elements
        if (s1.size == 1) & (s2.size == 1):
            likelihoods[idx - 1] = -np.inf
            continue

        # compute means
        mu1 = np.mean(s1)
        if s2.size != 0:
            mu2 = np.mean(s2)
        else:
            # Prevent numpy warning for taking mean of empty array
            mu2 = -np.inf

        # compute pooled variance
        variance = ((np.sum((s1 - mu1) ** 2) + np.sum((s2 - mu2) ** 2))) / (
            n_elements - 1 - (idx < n_elements)
        )
        std = np.sqrt(variance)

        # compute log likelihoods
        likelihoods[idx - 1] = np.sum(norm.logpdf(s1, loc=mu1, scale=std)) + np.sum(
            norm.logpdf(s2, loc=mu2, scale=std)
        )

    return likelihoods
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -