def compute_loss(self, outputs, labels, training=True): return cross_entropy_loss( outputs, labels["classes_id"], weight=labels.get("weight"), label_smoothing=self.params.get("label_smoothing", 0.0), training=training, )
def compute_loss(self, outputs, labels, training=True, params=None): if params is None: params = {} return cross_entropy_loss(outputs, labels["classes_id"], label_smoothing=params.get( "label_smoothing", 0.0), training=training)
def compute_loss(self, outputs, labels, training=True, params=None): if params is None: params = {} return cross_entropy_loss(outputs, labels["classes_id"], label_smoothing=params.get( "label_smoothing", 0.0), mode=tf.estimator.ModeKeys.TRAIN if training else tf.estimator.ModeKeys.EVAL)
def _compute_loss(self, features, labels, outputs, params, mode): return cross_entropy_loss(outputs, labels["classes_id"], label_smoothing=params.get( "label_smoothing", 0.0), mode=mode)
def _compute_loss(self, features, labels, outputs, params, mode): return cross_entropy_loss( outputs, labels["classes_id"], label_smoothing=params.get("label_smoothing", 0.0), mode=mode)