This MR is adding evalutation of confidence misscalibration. Confidence calibration is a topic investigated by i.e. Kato and Kato and others and is gaining relevace. Since its just adding an additional metric to the evaluation script its not to much for those not interested in this metric. Alternatively i could also disable it by a configuration in the config file.
This MR is adding evalutation of confidence misscalibration. Confidence calibration is a topic investigated by i.e. Kato and Kato and others and is gaining relevace. Since its just adding an additional metric to the evaluation script its not to much for those not interested in this metric. Alternatively i could also disable it by a configuration in the config file.