Add pythonic code structure

.gitignore

@@ -0,0 +1,5 @@
+uv.lock
+.venv
+data
+logs
+__pycache__

README.md

@@ -41,24 +41,33 @@ current implementation  is slightly different. We leverage predefined ResNet Mod
 
 #### Quick Start 
 
-copy the code
+You can either install directly from the repository:
+
+```bash
+pip install git+https://github.com/yhgon/NoProp.git
 ```
+
+Or download the repository and install it in editable mode:
+
+```bash
 git clone https://github.com/yhgon/NoProp.git
+cd NoProp
+pip install --editable .
 ```
 
 run with default option for mnist dataset
 ```
-python NoProp/src/nopropct_mnist.py
+noprop-mnist
 ```
 
 run with configure dataset and backbone model 
 ```
-python NoProp/src/noprop_simple.py --dataset cifar10 --backbone resnet50
+noprop-simple --dataset cifar10 --backbone resnet50
 ```
 
 run with configure dataset and backbone model  default epoch is `400`
 ```
-python NoProp/src/noprop_simple.py --dataset cifar100 --backbone resnet152
+noprop-simple --dataset cifar100 --backbone resnet152
 ```
 
 ## log 

pyproject.toml

@@ -0,0 +1,18 @@
+[build-system]
+requires = ["hatchling", "hatch-vcs"]
+build-backend = "hatchling.build"
+
+[project]
+name = "noprop"
+dynamic = ["version"]
+description = "Add your description here"
+readme = "README.md"
+requires-python = ">=3.10"
+dependencies = ["torch", "torchvision"]
+
+[project.scripts]
+noprop-simple = "noprop.simple:main"
+noprop-mnist = "noprop.mnist:main"
+
+[tool.hatch.version]
+source = "vcs"

src/nopropct_mnist.py → src/noprop/mnist.py

@@ -277,6 +277,9 @@ def train_and_eval(backbone: str):
     del model, optim, tr, te, ds_train, ds_test
     torch.cuda.empty_cache(); gc.collect()
 
-if __name__ == '__main__':
+def main():
     for backbone in ['resnet18', 'resnet50', 'resnet152']:
         train_and_eval(backbone)
+
+if __name__ == '__main__':
+    main()

src/noprop_simple.py → src/noprop/simple.py

@@ -393,7 +393,7 @@ def train_and_eval(backbone: str, time_emb_dim: int, embed_dim: int, dataset: st
     del model, optimizer, ds_train, ds_test, tr_loader, te_loader
     torch.cuda.empty_cache(); gc.collect()
 
-if __name__ == '__main__':
+def main():
     parser = argparse.ArgumentParser()
     parser.add_argument('--dataset', choices=['mnist','cifar10','cifar100'], required=True)
     parser.add_argument('--data-root', default='./data')
@@ -412,3 +412,6 @@ def train_and_eval(backbone: str, time_emb_dim: int, embed_dim: int, dataset: st
         data_root   = args.data_root,
         epoches     = args.epoches,
     )
+
+if __name__ == '__main__':
+    main()