ytm.guided.backends.act_loss.Tversky

class ytm.guided.backends.act_loss.Tversky(alpha: float = 0.5, beta: float = 0.5, eps: float = 1e-7)

Bases: ActLoss

Sigmoid with Tversky loss over the classes of a sample, for multi-label classification.

Parameters:

alphafloat, default=0.5

Weight of false positives, must be >= 0.

betafloat, default=0.5

Weight of false negatives, must be >= 0.

epsfloat, default=1e-7

Added to the numerator and denominator, must be > 0.