License is GPLv2

This is my experimentation with CIFAR-10 and Neural Networks with Pytorch and tools.

You'll need to install a few packages. You can try to use newer versions. Just omit the == below.

pip install torch==1.6.0
pip install torchvision==0.7.0
pip install matplotlib==3.2.2
pip install tensorboardX==2.1

You can use the code right out the gate to predicut. Just go into the cifarnn repo and run:

python predict <imagefile>

Where is the image file you want my already train neural Networks to guess.

If you want to re-train the model, run:

python train

Feel free to tweak parameters as you see fit. I did not specify any parameters to be passed in via argument for the train option, only because I preferred to tweak in code (or use optuna for hyperparameter calibration).

Nice things to do:

• Make it so predict doesn't require matplotlib, tensorboardX
• Get better than 92.623%