2024 Mini-project for Deep Learning
The best model reported in the paper is final_best_3.pth. See kaggle.ipynb or kaggle.py on how to load it in. Our highest score was 0.842.
Our best model architecture is in dropoutresnet.py. Our baseline model architecture is in modifiedresnet.py.
To see the results of our best model (96.10%), look at best_model.ipynb.
To train and use our best model, look at best_model_train.py or best_model_train.ipynb.
To see our hyperparameter tuning experiments, go check out our git branches.
All checkpoints were trained on HPC, so use NVIDIA GPU to run the notebooks and python files.