emukans / minDL

Minimalistic set of tools for Deep Learning.

minDL

Minimalistic set of tools for Deep Learning. The project is used as part of video course on YouTube. (English|Russian)

!NB, this repository is for educational purposes and is not intended for production use. For production use PyTorch, Tensorflow, etc.

Getting started

1. Create a new environment

   python -m venv venv

2. Activate the new environment (for Windows the activate script will have a different name)

   source venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

Neural network implementations

There are 2 implementations of a feedforward neural networks so far: sequential and vectorized.

plaindl - Implements the training algorithm on pure python. The purpose of this realisation is just to show, that there's no "magic" in the neural networks.

mindl - Implements the training algorithm using only numpy for matrix computations and code simplification.

Train your own network

You could either run the run.py script or start JupyterLab and check out more examples:

jupyter lab

If the Lab is not opened automatically, then open http://localhost:8888/lab