DeepTorchBlocks 0.0.1a0

DeepTorchBlocks

https://github.com/StefanoGiacomelli/DeepTorchBlocks/tree/main/nn_modules

...a repository for SoA Neural Networks building blocks (written in pure Python & PyTorch)
Read the API Documentation

View Demo · Report Bug · Request Feature

DeepTorchBlocks

This is a work-in-progress repository: please do not complain about code instability!

This project was born with the aim to collect and develop a general-purpose set of building blocks, for Neural Network architectures programming and researches structuring. For various reasons, one will be faced with the impossibility of directly exploiting pre-trained architectures and/or modules and will be forced to re-implement from scratch each layer (and or component) of the desired algorithm. To facilitate this step, DeepTorchBlocks gives us a hand with a careful selection (under continuous revisioning and updating) of state-of-the-art modules coming from CV, NLP and other reearch fields, inheriting all standard nn.Module (PyTorch) features and methods.

You will not find any pre-trained NNs architecture here, only building (sub-)modules!

(back to top)

Built With

For end-users:

• NumPy & related dependecies (for math operations)
• PyTorch & related dependecies (for modules inheritance)

For developers:

• torch-summary/torchinfo (for modules parameters report)
• torchview & related dependecies (for architectural monitoring)
• pytest & related dependecies (for modules testing)

...

(back to top)

Getting Started

This is an example of how you may set up your project locally. To get a local copy up and running follow these simple steps.

Cloning repository

• for Colab Notebooks

  !git clone https://github.com/StefanoGiacomelli/DeepTorchBlocks.git
  %cd './DeepTorchBlocks/nn_modules'
  

call Class and/or Functions directly from related Python files.

Installation

1. Install the package

   !pip install DeepTorchBlocks
   

   this will also auto-install required dependecies

2. Install auxiliary packages (for developers only)

   !pip install torchview
   from torchview import draw_graph
   
   !pip install torchinfo
   from torchinfo import summary
   

(back to top)

Usage

device = 'cuda' if torch.cuda.is_available() else 'cpu'
x = torch.rand([1, 3, 200, 200], device=device)              # An example RGB tensor of a single batch, with 200x200 pixels

model = ConvBlock(in_channels = 3,
                  out_channels = 10,
                  kernel_size = 3,
                  stride = 2,
                  padding = 1,
                  activation = nn.ReLU(),
                  device=device)

model(x)
>>> torch.Tensor(1, 10, 100, 100)

For more examples, please refer to the Documentation

(back to top)

Roadmap

• Add Bi-dir ConvRNN
• Define Tests pipeline (shape, self.params correctness, forward etc.)
• Compile documentation (SPhinx & ReadTheDocs template w. LaTeX extensions)
• Metrics.py (classification) & Utils.py (dataset, pre-processing
• CI/CD .github-workflow: for each admin push: Test-pipeline -> Package compilation & pubblication -> Documentation update

(back to top)

Contributing

Contributions are what make the open source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated. If you have a suggestion that would make this better, please fork the repo and create a pull request. You can also simply open an issue with the tag "enhancement". Don't forget to give the project a star! Thanks again!

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/NewFeature)
3. Commit your Changes (git commit -m 'Add some NewFeature')
4. Push to the Branch (git push origin feature/NewFeature)
5. Open a Pull Request

(back to top)

License

Distributed under the MIT License. See LICENSE.txt for more information.

(back to top)

Contact

Stefano Giacomelli - https://www.linkedin.com/in/stefano-giacomelli-811654135/ - stefano.giacomelli@graduate.univaq.it

Project Link: https://github.com/StefanoGiacomelli/DeepTorchBlocks

(back to top)

Acknowledgments

• Choose an Open Source License
• Img Shields
• GitHub Pages

(back to top)