majdjamal / deep_network

Deep Network implemented from scratch using only NumPy. This is my interpretation of Dense and Sequential available in the Tensorflow package.

Neural Network

Overview

Multilayer Perceptron with k Hidden Layers, built from scratch using NumPy. You can choose the number of hidden layers with the .Dense() method, a demonstration is found in experiments.py.

Hidden Layer values are computed with ReLU (eq. 1) and output probabilities with Softmax (eq. 2). The architecture is demonstrated in Figure 1.

(Figure 1. Architecture of a k-layer Neural Network)

Setup

Data

Required shape of the data is X.shape = (Ndim, Npts). Labels can have shape y.shape = (Npts,)

Environment

This model requires Python 3 and packages NumPy and Matplotlib.pyplot.

Demonstration

Navigate to the repository in your terminal and type:

python experiment.py

Questions

Feel free to send me your questions to [email protected].