Closed PoorvaRane closed 4 years ago
The kernel for PWL encodes the deltas between keypoint outputs (which are all positive for a monotonic layer). You need to do a cumsum over it, or simply use keypoints_outputs() methods on the layer: https://www.tensorflow.org/lattice/api_docs/python/tfl/layers/PWLCalibration#keypoints_outputs
I'm trying to train a calibration to incorporate into tf ranking, similar to https://github.com/tensorflow/lattice/issues/35
My code for the calibration looks like this:
The calibrations learned by this model are shown below. The bounds are being respected, but the calibrations do not appear monotonic. Any insights on what could be missing?