Neural Networks tutorial notes

This repository contains a multi-layer perceptron (3 layer fixed) implementation with experiments wrapped in a ipython notebook. Some notes and summaries were included as archive for personal learning purposes, however these notes are not necessary to reproduce the MLP implementation + experiments.

MLP implementation

The MLP is a 3 layer MLP that can learn both regression and classification problems. It has option to train using stochastic gradient descent and ensemble. Usage examples can be found inside the notebook.

The source code for this implementation is in mlp.py

Experiments in the notebook

3 experiments are performed in the notebook.

• XOR
• sine function over linear combination of 4 variables
• hand written letter prediction (engineered feature)

Different parameter configuration and their corresponding results are included for comparison