def __call__(self, y_pred: torch.Tensor, y_true: torch.Tensor) -> torch.Tensor:
y_pred = convert_to_tensor(y_pred)
y_true = convert_to_tensor(y_true)
check_same_shape(y_pred, y_true)
diff = y_pred - y_true
return torch.mean(torch.log((torch.exp(diff) + torch.exp(-1.0 * diff)) / 2.0))
def __call__(self, y_pred: torch.Tensor, y_true: torch.Tensor) -> torch.Tensor:
y_pred = convert_to_tensor(y_pred)
y_true = convert_to_tensor(y_true)
check_same_shape(y_pred, y_true)
squared_log = torch.pow(torch.log1p(y_pred) - torch.log1p(y_true), 2)
return torch.mean(squared_log)
def __call__(self, y_pred: torch.Tensor,
y_true: torch.Tensor) -> torch.Tensor:
y_pred = convert_to_tensor(y_pred)
y_true = convert_to_tensor(y_true)
check_same_shape(y_pred, y_true)
y_pred = torch.clamp(y_pred, self.epsilon, 1)
y_true = torch.clamp(y_true, self.epsilon, 1)
kld = torch.sum(y_true * torch.log(y_true / y_pred), axis=-1)
return torch.mean(kld)
def __call__(self,
y_pred: torch.Tensor,
y_true: torch.Tensor,
delta: float = 1.0) -> torch.Tensor:
y_pred = convert_to_tensor(y_pred)
y_true = convert_to_tensor(y_true)
check_same_shape(y_pred, y_true)
abs_error = torch.abs(y_pred - y_true)
quadratic = torch.clamp(abs_error, max=delta)
linear = abs_error - quadratic
loss = 0.5 * quadratic.pow(2) + delta * linear
return loss.mean()