Description: PET ML is a Python-based toolkit that utilizes input files to abstract common ML routines into something closer to a simulation framework to facilitate adoption by users outside the AI & ML community. The tool focuses on multiple-input multiple-out time series problems and vector-to-vector surrogate models. PET ML utilizes an encoder-decoder framework that enables modular model designs for a series of tasks; users can select model components and settings to design a wide variety of architectures. Wrapper classes enable simplified training of the models for a range of tasks and functionalities. These support two means of uncertainty quantification: Monte Carlo dropout and conformalized quantile regression.

CUI