Add pinball loss

[RomainBrault]

Describe the feature and the current behavior/state.

The pinball loss is a very simple modification of the MAE and allows (conditional) quantile regression. This is rather simple to code, and would be of interest for a large community.

This has been submitted to TensorFlow (tensorflow/tensorflow#32871) and redirected here.

A pull request is open (#579)

Relevant information

• Are you willing to contribute it (yes/no): yes
• Are you willing to maintain it going forward? (yes/no): yes
• Is there a relevant academic paper? (if so, where): yes, the original paper and many others
• Is there already an implementation in another framework? (if so, where): yes, some blogs propose implementations (https://towardsdatascience.com/deep-quantile-regression-c85481548b5a)
• Was it part of tf.contrib? (if so, where): no

Which API type would this fall under (layer, metric, optimizer, etc.)

metrics and losses (Keras)

Who will benefit with this feature?

People in econmics, biology. To model data uncertainty / distribution.

Any other info.

[Sandy4321]

do you have full example for multivariate quantile regressing with using model = tf.keras.Model() model.compile('sgd', loss=tfa.losses.PinballLoss(tau=.1))

from https://www.tensorflow.org/addons/api_docs/python/tfa/losses/PinballLoss