To download the repository locally, please download the file PEPITA.zip
- 1 Hidden layer:
python scripts/train.py -en test_1 -d cifar100 -ls 1024 -wd 0.0 -bi uniform -li he_uniform - 2 Hidden layers:
python scripts/train.py -en test_2 -d cifar100 -ls 1024 1024 -wd 0.0 -bi uniform -li he_uniform - 3 Hidden layers:
python scripts/train.py -en test_2 -d cifar100 -ls 1024 1024 1024 -wd 0.0 -bi uniform -li he_uniform
Run python scripts/train.py -h to have an overview of the flags and their usage.
All experiments are logged by tensorboard in the experiments folder.
The following libraries are required for running the code. The indicated versions are the one we used, but other versions may work.
torch==1.13.1
torchvision==0.14.1
pytorch_lightning==1.6.5
torchmetrics==0.9.2
loguru==0.6.0
numpy==1.21.2
pillow==8.4.0
scikit-learn==1.2.1
Additionally, ray[tune] is needed for grid searches.