Neural Networks Project -- SemiTime

This repository contains the Project of the Neural Networks course of the master in Artificial Intelligence and Robotics at La Sapienza University in Rome.

The project has been based on this paper: Semi-supervised time series classification by temporal relation prediction (https://haoyfan.github.io/papers/SemiTime_ICASSP2021.pdf), which describes a novel approach for time series classification. In this work the authors have proposed a method of semi-supervised time series classification architecture (termed as SemiTime) by gaining from the structure of unlabeled data in a self-supervised manner.

Schematic illustration of semi-supervised techniques described	SemiTime architecture

Requirements

• Python 3.9.6
• torch 1.13.1
• torchvision 0.14.1
• numpy 1.24.1
• sklearn 1.2.1
• pandas 1.5.3

Run model

Training

Supervised:

cd Code
python main.py --dataset CricketX --task supervised --run train --save true

Semi-Supervised:

cd Code
python main.py --dataset CricketX --task semi-supervised --run train --save true

After training the model is saved in 'checkpoints' folder.

Test

The model is automatically tested after training. If you want to test it later:

Supervised:

cd Code
python main.py --dataset CricketX --task supervised --run test

Semi-Supervised:

cd Code
python main.py --dataset CricketX --task semi-supervised --run test

or you can also use the provided notebook.

Parse options:

• dataset: {CricketX, UWaveGestureLibraryAll, InsectWingbeatSound, MFPT, XJTU, EpilepticSeizure}
• task: {supervised, semi-supervised}
• run: {train, test}
• save: {true, false} (save or not the model in ./Code/checkpoint/{task}/{dataset})