A "from scratch" implementation of classic feed-forward neural networks for binary/multi-class classification using ReLU activations, cross entropy loss and sigmoid/softmax output.
Read through the documentation in neuralnet.py for a description of the
implementation.
An example usage of neuralnet.py is given in the Usage.ipynb notebook.
Alternatively you can open the whole code in Google Colab -> here.
neuralnet.py is purely implemented in NumPy:
- NumPy:
pip install numpy
To run the examples in the Usage.ipynb notebook locally please install the
requirements noted in requirements.txt:
- Requirements:
pip install -r requirements.txt
The following datasets are used in the examples:
- Multi-class classification: MNIST
- Binary-class classification: Breast Cancer Wisconsin (Diagnostic) Data Set