tf_agents.bandits.policies.loss_utils.pinball_loss

Adds a Pinball loss for quantile regression.

tf_agents.bandits.policies.loss_utils.pinball_loss(
    y_true: tf_agents.typing.types.Tensor,
    y_pred: tf_agents.typing.types.Tensor,
    weights: tf_agents.typing.types.Float = 1.0,
    scope: Optional[Text] = None,
    loss_collection: tf.compat.v1.GraphKeys = tf.compat.v1.GraphKeys.LOSSES,
    reduction: tf.compat.v1.losses.Reduction = tf.compat.v1.losses.Reduction.SUM_BY_NONZERO_WEIGHTS,
    quantile: float = 0.5
) -> tf_agents.typing.types.Float

```

loss = quantile * (y_true - y_pred) if y_true > y_pred loss = (quantile - 1) * (y_true - y_pred) otherwise ```

See: https://en.wikipedia.org/wiki/Quantile_regression#Quantiles

weights acts as a coefficient for the loss. If a scalar is provided, then the loss is simply scaled by the given value. If weights is a tensor of size [batch_size], then the total loss for each sample of the batch is rescaled by the corresponding element in the weights vector. If the shape of weights matches the shape of predictions, then the loss of each measurable element of predictions is scaled by the corresponding value of weights.

y_true tensor of true targets.
y_pred tensor of predicted targets.
weights Optional Tensor whose rank is either 0, or the same rank as labels, and must be broadcastable to labels (i.e., all dimensions must be either 1, or the same as the corresponding losses dimension).
scope The scope for the operations performed in computing the loss.
loss_collection collection to which the loss will be added.
reduction Type of reduction to apply to loss.
quantile A float between 0. and 1., the quantile we want to regress.

Weighted Pinball loss float Tensor. If reduction is NONE, this has the same shape as labels; otherwise, it is scalar.

ValueError If the shape of predictions doesn't match that of labels or if the shape of weights is invalid. Also if labels or predictions is None.

eager compatibility

The loss_collection argument is ignored when executing eagerly. Consider holding on to the return value or collecting losses via a tf.keras.Model.