spine.utils.weighting

Module which contains methods to compute class weights.

All methods compute the weights based on the relative abundance of each class.

Functions

get_class_weights(labels, num_classes[, ...])

Computes class-wise weights based on their relative abundance.
spine.utils.weighting.get_class_weights(labels, num_classes, mode='const', per_class=True)[source]

Computes class-wise weights based on their relative abundance.

Parameters:

  • labels (Union[np.ndarray, torch.Tensor]) –

    1. Array of class values to base the weighting scheme on

  • num_classes (int) – Total number of classes (needed if there are some missing)

  • mode (str, default const) – Weigthing mode (one of ‘const’, ‘log’, ‘sqrt’

  • per_class (bool, default True) – If True, returns one valuer per class. Otherwise, this function returns one value per input in the classes array