CIFAR-10-CNN

Just a simple 8 layer CNN developed to work on CIFAR-10. The code is inspired by pytorch mnist tutorial, ie https://github.com/pytorch/examples/blob/master/mnist/main.py as well as work from CS598 D from UIUC, which is a Deep Learning course I am taking.

Just a fun little project to show case some basic concepts like max pooling, batch normalization, and drop out.

Getting Started

To get started on the project is very easy, just

git clone [email protected]:yumochi/CIFAR-10-CNN.git

Prerequisites

To run the code you will need the following:

Python

(Refer to https://www.python.org/downloads/)

Pytorch

# Python 3.x
pip3 install torch torchvision

# Python 2.x`
pip install torch torchvision

(Refer to https://pytorch.org/get-started/locally/ for more info.)

Torchvision

pip install torchvision

(Refer to https://pypi.org/project/torchvision/0.1.8/ )

h5py (Not necessary)

pip install h5py

h5py was originally used to import image data, but the code is adopted to use Torchvision

Comment out code if not needed

Running the tests

To run the code, just run

python main.py

Set parameters with argparser

For a list of terminal commands for the argparser, refer to texts below or check in hw3.py for all parameters

Set batch-size with --batch-size x

x has to be an integer

python hw3.py --batch-size 16

Set epoch number with --epochs x

x has to be an integer

python hw3.py --epochs 30

Set learning rate with --lr x

x has to be an float

python hw3.py --lr 0.0001

Set sample number in monte carlo approximation with --mck x

x has to be an integer

python hw3.py --mck 16

Versioning

We use SemVer for versioning. For the versions available, see the tags on this repository.

Authors

• Yumo Chi - Initial work - CIFAR-10-CNN

Acknowledgments

• Pytorch Developers
• UIUC CS598D's professor and tas.