TorcHood

torchood: A User-Friendly Wrapper for Torch, Making Neural Network Training Effortless
Jumpstart your deep learning endeavors and rapidly prototype custom projects with torchood 🚀⚡🔥

├───torchood
│   │   dataset.py
│   │   trainer.py
│   │   __init__.py
│   │
│   ├───models
│   │   │   common.py
│   │   │   custom_resnet.py
│   │   │   mini_resnet.py
│   │   │   resnet.py
│   │   │   __init__.py
│   │
│   ├───utils
│   │       gradcam.py
│   │       misc.py
│   │       plotting.py
│   │       __init__.py

# clone project
git clone https://github.com/anantgupta129/TorcHood.git
cd TorcHood

# [OPTIONAL] create conda environment
conda create -n myenv python=3.10
conda activate myenv

# install pytorch according to instructions
# https://pytorch.org/get-started/

! python setup.py sdist
! pip install .

• LR finder

  from torch.nn import CrossEntropyLoss
  from torchood.utils.misc import find_lr
  
  optimizer = torch.optim.SGD(model.parameters(), lr=0.01, momentum=0.9)
  
  kwargs = {"end_lr":10, "num_iter": 200, "step_mode":"exp"}
  find_lr(model, device, optimizer, criterion=CrossEntropyLoss(), dataloader=train_loader, **kwargs)

• grad cam

  from torchood.utils.gradcam import plot_cam_on_image
  
  mean = (0.49139968, 0.48215841, 0.44653091)
  std = (0.24703223, 0.24348513, 0.26158784)
  plot_cam_on_image(model, [model.layer4[1]], imgs_list, {"mean": mean, "std": std})

• Training & training History

  from torchood.trainer import Trainer
  trainer = Trainer(model, device, optimizer, scheduler)
  for epoch in range(1, num_epochs + 1):
      print(f"Epoch {epoch}")
      trainer.train(train_loader)
      trainer.evaluate(test_loader)
  
  trainer.plot_history()

• Supports the CIFAR10 dataset as a sample. In the future, we plan to add support for additional datasets.

• Supports sample models

Contributions are invited! Don't hesitate to submit a pull request.