lstmix - V2.0

Version 2.0

Adding small features (e.g., weight initialization), fixes around the training algorithm (e.g., dynamic teacher forcing update based on check_epochs).

Model deployment from Pytorch and training algorithm has been heavily refactored to provide better encapsulation and maintainability throughout the framework. LSTM and GRU classes have now been re-factored as child classes from architecture type parents (RNNFC and RNNED), which at the same time are childs of a abstract base class General_RNN, which holds common functionalities between all RNN models deployed here. This allowed to remove separate ED and FC training functions, improving the code clarity and overall structure. These changes have been applied accordingly to all other functionalities dependent on model train (e.g., hyperparameter tuning, rollout, etc).

- Python
Published by jpv219 almost 2 years ago

lstmix - V1.0

First release after full implementation of the RNN framework employing GRU and LSTM cells. This version has been used to generate the results published in the special issue Multiscale Modelling and Artificial Intelligence for Multiphase Flow Science in the ACS journal Industrial Engineering & Chemistry Research.

- Python
Published by jpv219 almost 2 years ago