tf.keras.losses.logcosh(
y_true,
y_pred
)
Defined in tensorflow/python/keras/losses.py.
Logarithm of the hyperbolic cosine of the prediction error.
log(cosh(x)) is approximately equal to (x ** 2) / 2 for small x and
to abs(x) - log(2) for large x. This means that 'logcosh' works mostly
like the mean squared error, but will not be so strongly affected by the
occasional wildly incorrect prediction.
Arguments:
y_true: tensor of true targets.y_pred: tensor of predicted targets.
Returns:
Tensor with one scalar loss entry per sample.