Deep Learning Project - 2023/24
This repository is a template repository for the homeworks to be developed in the Deep Learning course.
Deep Learning is a course of the
- Master Degree in Computer Engineering of the Department of Information Engineering, University of Padua, Italy.
| NAME | SURNAME | |
|---|---|---|
| Francesco | Chemello | francesco.chemello.1@studenti.unipd.it |
| Pietro | Volpato | pietro.volpato@studenti.unipd.it |
The repository is organised as follows:
- code: folder that contains all the code for the homework + files for git.
- presentation: this folder contains the final presentation of the course.
- slides: this folder contains the slides used for presenting the project.
- developing: contains the files that the group uses for developing the application.
- Retrieve data from .csv files (training, validation and test set).
- OneHot coding from alphabetic to numeric (4xN matrix, N = sting size).
- Reshape the PyTorch tensor to fit the data.
- Implementation of CNN
- Implementation of LSTM
- Training the Nets implementing the early stop
- Run on the test set.
- Implementation of the ensemble learning for choosing the best label.
- Performance evaluation: metrics used by ViraMiner.
- Discussion of results + ppt presentation of work or pdf (as directed by the professor).
The code is written in .ipynb and before start the run is better to set the hyperparameters that are avaiable on 13th code cell.
These are:
- path to the dataset.
- 1 for data from local machine: the data must be inside a folder called data inside the folder where there is the code.
- 2 for data from Colab: the data must be inside in the default folder.
- 3 for data from Google Drive: the data must be inside the Colab folder.
- number of data to be upsample.
- number of data to be generated by SMOTE.
- type of RNN.
- 1 for GRU.
- 2 for LSTM.
Note: if you want to test the code in your local machine you need to have at least 8 Gb of RAM and 6 GB of GRAM (if you have cuda) or at least 16 Gb RAM [1].
[1] data taken during Colab simulations.
All the contents of this repository are shared using the Creative Commons Attribution-ShareAlike 4.0 International License.
