To run HN use the following command:
python3 testac.py --dir=<DIR> --dataset=CIFAR100 --data_path=<PATH> --model=PreResNet164 --epochs=225 --lr_init=0.1 --wd=3e-4 --swa --swa_start=126 --swa_lr=0.05
Parameters:
DIR
โ path to training directory where checkpoints will be storedPATH
โ path to the data directory
- Test accuracy (%) of HN on CIFAR-100 on PreResNet164 (150) is 80.4
- Our model has 1.7M parameters.
To train HN use the following command:
python3 train.py --dir=<DIR> --dataset=CIFAR100 --data_path=<PATH> --model=PreResNet164 --epochs=225 --lr_init=0.1 --wd=3e-4 --swa --swa_start=126 --swa_lr=0.05
We use the SWA method to train our model.