MO346/MC934 Course work

This repository contains code and instructions to support the MO346/MC934 course work project.

This project has the following structure:

• configs/: contains the configuration files to instantiate models, data modules, and trainers. It is used by LightningCLI to instantiate experiments. The configuration files are written in YAML format and contain the class path (package and class name) and parameters to instantiate the objects.
• data/: contains the data used in the project. Your data should be stored here. This directory is not tracked by git.
• data_modules/: contains implementations of LightningDataModule classes. These classes are responsible for loading and preprocessing data, and also for splitting the data into training, validation, and test sets.
• models/: contains implementations of LightningModule classes. These classes are responsible for defining the model architecture and implementing the forward, training_step, and configure_optimizers methods.
• transforms/: contains implementations of Numpy/PyTorch transforms. These transforms are used to preprocess data before feeding it to the model.

The main.py file is the entry point of the project. It is a CLI that is responsible for instantiating the trainer, model, and data module, and also performing the training and testing of the model. Configuration files are used to instantiate each of these objects. The configuration files are stored in the configs/ directory.

For instance, to train and test the model, you can run the following command:

python main.py --trainer configs/trainer/default.yaml --model configs/models/mlp.yaml --data configs/data_modules/har.yaml

The --trainer contains the class and parameters for instantiating a Lightning.Trainer object. The --model contains the name of the class and parameters for instantiating a LightningModule object. The --data contains the name of the class and parameters for instantiating a LightningDataModule object. Output files are stored in the logs/ directory. This directory is not tracked by git.

The trainer will use standard fit and test methods from the LightningModule class to train and test the model. The fit method is used to train the model, and the test method is used to test the model.

Tutorial notebooks

The 1-HAR-dataloader-exploration.ipynb and 2-HAR-MLP-training.ipynb notebooks contains instructions on how to read the HAR dataset and train an MLP model to solve the HAR task. The 3-Seismic-dataloader-exploration.ipynb and 4-Seismic-Deeplab-training.ipynb notebooks contains instructions on how to read the Seismic dataset and train an Deeplab v3 model to solve the Seismic task.

Installation

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.txt

Running the tests

To run the tests, you can run the following command:

run.sh

Adding new features

To add new features to the project, you can create new files in the data_modules/, models/, and transforms/ directories. Once you have created the new files, you can create new configuration files in the configs/ directory.

Authors

• Otávio Napoli
• Edson Borin
• Gabriel Gutierrez

License

This project is licensed under GPL-3.0 License - see the LICENSE file for details.