Losses API

Auto-generated from source — a multi-backend Keras port of torchgeo.losses. See Data for how these fit into a training loop.

Losses

keras_climate.losses.elects.EarlyRewardLoss

EarlyRewardLoss(alpha=0.5, epsilon=10, weight=None)

ELECTS loss for early classification of time series.

Source code in keras_climate/losses/elects.py

def __init__(self, alpha=0.5, epsilon=10, weight=None):
    if not 0 <= alpha <= 1:
        raise ValueError(f"Invalid alpha value: {alpha}")
    if not 0 <= epsilon:
        raise ValueError(f"Invalid epsilon value: {epsilon}")

    self.alpha = alpha
    self.epsilon = epsilon
    self.weight = None if weight is None else ops.convert_to_tensor(weight)

keras_climate.losses.qr.QRLoss

QRLoss(eps=1e-08)

The QR (forward) loss between class probabilities and predictions.

Source code in keras_climate/losses/qr.py

def __init__(self, eps=1e-8):
    if not 0.0 <= eps:
        raise ValueError(f"Invalid epsilon value: {eps}")
    self.eps = eps

keras_climate.losses.qr.RQLoss

RQLoss(eps=1e-08)

The RQ (backwards) loss between class probabilities and predictions.

Source code in keras_climate/losses/qr.py

def __init__(self, eps=1e-8):
    if not 0.0 <= eps:
        raise ValueError(f"Invalid epsilon value: {eps}")
    self.eps = eps