This repository contains code and instructions to support the MO346/MC934 course work project.
This project has the following structure:
data/: contains the data used in the project. Your data should be stored here. This directory may contain large datasets and must not be added to the Git repository.data_modules/: contains implementations ofLightningDataModuleclasses. These classes are responsible for loading and preprocessing data, and also for splitting the data into training, validation, and test sets.models/: contains implementations ofLightningModuleclasses. These classes are responsible for defining the model architecture and implementing theforward,training_step, andconfigure_optimizersmethods.transforms/: contains implementations of Numpy/PyTorch transforms. These transforms are used to preprocess data before feeding it to the model.best_results/: contains the scripts to pretrain the backbone, train the downstream model, and evaluate the downstream model with the best hyperparameters (e.g., learning rate) for each technique. The scripts in this directory must follow the API defined by the professor.report_results/: contains the scripts to generate the results presented at the technical report.
The ipynb files at the root directory (e.g., 1-HAR-dataloader-exploration.ipynb, 2-HAR-MLP-training.ipynb, etc.) are tutorial notebooks with instructions on how to read the HAR/seismic dataset and train ML models to solve the HAR/seismic tasks.
We use VSCode Containers to run the project. To run the project, you need to have Docker and VSCode installed on your machine. If you don't know how to use Containers in VSCode, you can follow the instructions in the following link.
Once inside the container, you can install the dependencies, running the following command:
pip install -r requirements.txtTo run the example code, you must change into the best_results/seismic/ directory and run the following commands:
python BT_pretrain.py
python BT_train.py
python BT_evaluate.pyThis project is licensed under GPL-3.0 License - see the LICENSE file for details.