Functional Neural Networks

Two folders: scalar-on-function and function-on-function are used to store files related to different tasks. Each of the folders contains the following files:

• Real - real datasets and corresponding scripts used for fine-tuning and producing the results.

• Simulation - folder data used to generate the data and the corresponding scripts used to conduct the simulation experiments.

• Models.py - all the PyTorch neural network model classes

• Utils.py - standard FDA models, PyTorch helper functions, cross-validation, etc.

  Inside the folders Real and Simulation are the folders with task names e.g. Tecator, Bike Sharing, task 2, etc. that contain the datasets, files facilitating the fine-tuning process via ray library and final scripts (with the names corresponding to the task e.g. Tecator dataset -> Tecator.py) that were hard-coded with the results obtained during the fine-tuning. During the fine-tuning of each model, a corresponding auxiliary function is imported from train_functions.py. This process was needed because of different data formats, number of functional inputs, basis selection, etc.