This repository contains starter code to use when crafting your solution. You should clone it on your machine and work from there.

• model.py is where you define your model
• prep_data.py is used to process the data for training
• train.py contains the code for training your model
• test.py contains a few useful functions to evaluate your solution as well as the cross-validation testing we will use to pick winners.
• main.py outputs information about the data, your model architecture, and your model performance. Run this file and include a screenshot of the output in your DevPost submission.

1. Make any desired alterations to the data, and update prep_data.get_prepared_data() to return your version of the data if needed.
2. Create your model (change the MyModel class in model.py).
3. Test your model (using your own code, or one of the provided functions in train.py/test.py).
4. Repeat steps 1-3 until you have achieved the highest accuracy possible across the entire dataset.
5. Run main.py and include its output in your submission.

• Your model must take the form of a class in model.py with the name MyModel that extends torch.nn.Module.
• Do not include any data or model weights in your repository on GitHub.
• Do not modify main.py in any way.
• We should be able to reproduce your stated results exactly by downloading the datasets you cited in your DevPost and placing them in the existing data folder, then running main.py. Make sure you haven't manually renamed any dataset files.
• You may use any open source model as a component of your solution (not your entire solution). Your code must download existing models itself, without requiring API keys or authentication of any kind.
• Your model must be reasonably efficient. If it takes more than 10 minutes to train your model on a single A100 GPU, or the model requires more than 4GB of GPU memory, you will be disqualified.
• You may not commit to your repository after 11:59pm on Monday, February 10th, 2025.