This repository contains the material related to the 2nd homework of the Artificial Neural Networks and Deep Learning Course, Politecnico di Milano.
Students: Nicola Farenga, Giorgio Longari.
The folder data-exp-preprocessing contains two notebooks concerning data exploration, preprocessing and loading.
The architectures that have been tested are the following:
- Base models (SimpleRNN, LSTM, GRU)
- LSTM Seq2Seq
- LSTM Seq2Seq + Attention
- Dilated causal CNN + LSTM (WaveNet-style)
They are all listed in the models.ipynb notebook and the notebooks referred to their evaluation and comparisons can be found in the
architectures-evaluation folder.
We have performed multiple tests under different hyperparameters combinations, they can be found in the hyperparameters-tuning folder.
After selecting the best models we have retrained them on the complete dataset, the notebooks are stored in the final-training folder.