Bishõgrad

A Bishõ Autograd engine in python along with a lightweight deep neural-network library! (inspired from mircrograd by Andrej Karpathy)

what's Bishõ?

微小 - Bishõ is Japanese word for 'tiny' since my implementation is very tinyyyyy ^_^ compared to PyTorch/Tensorflow

what's Hako?

箱 - Hako is Japanese word for 'box' , here Hako signify the neurons in our network ;D

Installation

1. Manually build the wheel file and install it:

   • Clone this repo:

     git clone https://github.com/AK3847/Bishograd.git
     cd Bishograd
     

   • Use a python package manager like Rye to build the wheels:

     rye build
     

   • Use pip to install the package into your virtual env:

     pip install path/to/the/wheel/file/bishograd-0.1.0-py3-none-any.whl
     

2. Download the temporary wheel file from here

   • Use pip to install the package into your virtual env:

     pip install path/to/the/wheel/file/bishograd-0.1.0-py3-none-any.whl
     

Example

Wonder how this works? Checkout the examples.

Targets :

• ReLU Activation function
• Add MLP.training() to automate the whole training code
• Add Sigmoid, LeakyReLU & other activation functions
• Add loss functions - categorical loss, mean-square loss etc

Contribution :

This repository is open for contribution! Primary way to contribute is to either raise an issue or a pull request with proper description and code format. You can contribute to any Targets or add new features as well.