Documentation of DQS

DQS is a neural network toolkit for distribution regression, quantile regression, and survival analysis. This toolkit provides various classes and methods useful for predicting probability distribution. This toolkit is currently available for PyTorch.

How to install

pip install dqs

Overview:

• How to Use
• Predicting Probability Distribution
• How to use dqs.torch.loss.NegativeLoglikelihood

API Reference:

• dqs.torch.distribution.DistributionLinear
• dqs.torch.layer.HierarchicalSoftmax
• dqs.torch.layer.SigSoftmax
• dqs.torch.loss.Brier
• dqs.torch.loss.CensoredBrier
• dqs.torch.loss.CensoredNegativeLoglikelihood
• dqs.torch.loss.CensoredRankedProbabilityScore
• dqs.torch.loss.NegativeLogLikelihood
• dqs.torch.loss.Pinball
• dqs.torch.loss.Portnoy
• dqs.torch.loss.RankedProbabilityScore

Citation

Please consider citing this paper: H. Yanagisawa, “Proper Scoring Rules for Survival Analysis,” ICML 2023 (to appear).