Final Project Scripts - SYS819

Note: The scripts used in this project are only designed to work with GPUs if Apple Silicon is present.

Virtual environment

Use conda to create new virtual environment:

conda create --name sys819_project

Activate virtual environment:

conda activate sys819_project

Install Python at version 3.12 to ensure compability with Pytorch libraries:

conda install python=3.12

Next, install Pythorch libraries:

pip install torch torchvision

Then install timm (library that has various pre-trained models):

pip install timm

Then install matplotlib to plot graphs:

pip install matplotlib

Hyperparameter tuning script

You can run the Python file train_hyperparameter_tuning.py under the virtual environment by making sure the environment is active and then:

python train_hyperparameter_tuning.py --model tiny_vit_21m_224

To find a set of optiomal hyperparameters for TinyViT-21M, or:

python train.py --model resnet34d

To find a set of optiomal hyperparameters for ResNet-34.

Train script setup

You can run the Python file train.py under the virtual environment by making sure the environment is active and then:

python train.py --model tiny_vit_21m_224

To train TinyViT-21M, or:

python train.py --model resnet34d

To train ResNet-34.

Plot script setup

You can run the Python file train_plots.py under the virtual environment by making sure the environment is active and then:

python train_plots.py