TURegCheckpoint#

class torch_uncertainty.callbacks.TURegCheckpoint(probabilistic=False, save_last=False)[source]#

Keep multiple checkpoints corresponding to the best model in terms of: Mean Squared Error, and eventually the Negative Log-Likelihood and Quantile Calibration Error.

Parameters:
  • probabilistic (bool, optional) – If True, also tracks the Negative Log-Likelihood and the Quantile Calibration Error. Default to False.

  • save_last (bool | "link", optional) – When True, saves a last.ckpt copy whenever a checkpoint file gets saved. Can be set to "link" on a local filesystem to create a symbolic link. This allows accessing the latest checkpoint in a deterministic manner. Default to False.