I am interested in calibrating a binary probabilistic classifier in TFX. I was about to try doing it in standard Python externally to TFX, but then I found this piecewise linear calibration layer.
The description is a bit cryptic to me. Is this layer the sort of thing one could stack to the output layer of a TFX model and calibrate the output using recent y_true and y_pred?
If not, is there a standard way to do calibration in TFX?
Calibration of the data should be done prior to the data the transformation and classification.
The piecewise data is only applicable when the data coincides to regions of observed data. We are not given enough information to properly answer this question.