This is a project to make a simple, encapsulated and well commented and documented framework for making neural nets. As far as possible, I adhere to OOP paradigms and intend the library as a soft introduction to neural networks.
You will need to install some dependencies for the code to run. These are minimal as code is mostly just vanilla python, NumpPy and graphing libraries. These can be installed by running
pip install -r requirements.txt
A write up of how the library is structured and some relevant mathematics can be found here
Some examples solved using the library can be found here
Documentation is auto-generated using PyDoc. Navigate to AI-Library directory and run
pydoc -b
to open server on an arbitrary unused port.
Layer implementations for:
- Dense layer
- Convolutional Layer
- Reshape layer
- Recurrent neural network (not a layer but don't have anywhere else to mention it)
Activation function implementations for:
- Tanh
- Logistic
- RelU
- Softmax
Loss function implementations for:
- Mean Squared Error
- Binary Cross Entropy
- Categorical Cross Entropy
- Sum of Squared Errors
- Documentation (as of 12/07/24)
- Writeup (as of 14/07/24)