reproduce.sh

#/bin/bash

LR=0.1
BS=128
DROPOUT=0.5
WD=1e-4

## MNIST
COMMON="--filters 64 --dataset mnist --augmentation none --epochs 25 --batch-size $BS --lr $LR --patience 5 --dropout $DROPOUT --wd $WD"

python train.py --model resnet --downsample residual $COMMON
python train.py --model odenet --downsample residual --adjoint $COMMON
python train.py --model odenet --downsample one-shot --adjoint $COMMON

find runs_mnist -maxdepth 1 -mindepth 1 -exec python evaluate.py nfe {} --tol 0.001 \;
find runs_mnist -maxdepth 1 -mindepth 1 -exec python evaluate.py features {} \;
find runs_mnist -maxdepth 1 -mindepth 1 -exec python evaluate.py retrieval {} \;


## CIFAR-10
COMMON="--filters 256 --dataset cifar10 --augmentation crop+jitter+flip+norm --epochs 250 --batch-size $BS --lr $LR --patience 15 --dropout $DROPOUT --wd $WD"

python train.py --model resnet --downsample residual $COMMON
python train.py --model odenet --downsample residual --adjoint $COMMON
python train.py --model odenet --downsample one-shot --adjoint $COMMON

find runs_cifar10 -maxdepth 1 -mindepth 1 -exec python evaluate.py nfe {} --tol 0.001 \;
find runs_cifar10 -maxdepth 1 -mindepth 1 -exec python evaluate.py features {} \;
find runs_cifar10 -maxdepth 1 -mindepth 1 -exec python evaluate.py retrieval {} \;