https://github.com/utility-code/tinydl

Tiny, easy to understand deep learning library in Python. With GPU support and only in numpy :)
https://github.com/utility-code/tinydl

deep-learning easy-to-understand gpu library numpy python

Last synced: about 1 month ago
JSON representation

Tiny, easy to understand deep learning library in Python. With GPU support and only in numpy :)

• Host: GitHub
• URL: https://github.com/utility-code/tinydl
• Owner: utility-code
• Created: 2021-03-15T10:46:44.000Z (almost 4 years ago)
• Default Branch: master
• Last Pushed: 2021-04-27T12:26:46.000Z (over 3 years ago)
• Last Synced: 2023-03-05T03:09:30.502Z (almost 2 years ago)
• Topics: deep-learning, easy-to-understand, gpu, library, numpy, python
• Language: Python
• Homepage:
• Size: 433 KB
• Stars: 16
• Watchers: 2
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# TinyDL

- Tiny Deep learning library 

- [DOCUMENTATION](https://subhadityamukherjee.github.io/tinyDL/)

- Check the demo.py or the demo notebook!!! (same content)

- Have a look at what works so far :)

- Only needs numpy and pycuda for the library

## Contributions

- Do check the todo.md for stuff you can work on :)

- Welcome! Add an issue if you have something you want to look at

- PRs welcome

## Requirements

- numpy

- matplotlib

- pycuda  (for GPU)

- pandas (if you are using the helpers and need to read a table)

## How to run

- Install the requirements

- Configure parameters in config.py

- python demo.py

## Why?

- Pytorch is too complicated to learn from. (Please. I tried. There are a million folders. My brain >.<)

- This does not intend to be Pytorch. Just to understand what goes into it

- An attempt at recreating most of the essential components from scratch

- Eventual blogs on it as well

## Inspired by

- Karpathy and his awesomeness xD

- [micrograd](https://github.com/karpathy/micrograd)

- [teddykokker](https://github.com/teddykoker/tinyloader)

- ft. Every attempt of me trying this in other languages and failing miserably ):

## What are the .sh files

- I am lazy. So I write code when I have to do the same things again and again.

### pusher.sh

- Formats all the code using "Black" formatter

- Creates documentation from docstrings in the code using pandoc

- Fixes the documentation paths for working with Github Pages

- Takes the demo.py and pops it into a nice notebook for anyone to run and use

- If an argument is given, it git commits with the message and pushes it to the repository

## References

- [training and tensor operations](https://github.com/kartik4949/deepops)

- [autograd](https://github.com/karpathy/micrograd)

- [pbar](https://stackoverflow.com/questions/3173320/text-progress-bar-in-the-console)

- [skalski](https://github.com/SkalskiP/ILearnDeepLearning.py/blob/master/01_mysteries_of_neural_networks/03_numpy_neural_net/Numpy%20deep%20neural%20network.ipynb)

- [softmax](https://stackoverflow.com/questions/34968722/how-to-implement-the-softmax-function-in-python)

- [activations1](https://machinelearningmastery.com/weight-initialization-for-deep-learning-neural-networks/)

- [albumentations](https://albumentations.ai/docs/api_reference/)

- [conv](https://medium.com/analytics-vidhya/2d-convolution-using-python-numpy-43442ff5f381)

- [optim](https://github.com/ilguyi/optimizers.numpy/blob/master/02.stochastic.gradient.descent.ipynb)

- [loss](https://github.com/Nico-Curti/NumPyNet/blob/master/NumPyNet/metrics.py)

- [loss2](https://github.com/ddbourgin/numpy-ml/blob/master/numpy_ml/neural_nets/losses/losses.py#L514-L924)

- Countless stackoverflow searches (hehe)