Developer Docs#
The following are ideas for future PES-Learn development
Feature Ideas: - Using PES-Learn to compute trajectories of PES’s - Add more sampling methods - Scale up to global PES to look at conformers - Add GPU support - Add threading with PyTorch - Add more transformations other than Morse transforms - Use the surface as a program to find features such as global minima (e.g. with gradient descent) - User specified hyperparameter optimization bounds - User designed models - Interactive graphing - Include data reduction keyword - Index training and test set corresponding to original PES.dat - Diabatization incorporation (e.g. DDNN of Truhlar) - Active learning for larger systems - Prune Sobol’ sampling to prevent bunched up data points