vvkmnn / introdl Goto Github PK

The introductory project for the Deep Learning Foundations Nanodegree program, designed to introduce the concepts of Neural Networks. This is technically not my first neural network, but it was still a really fun project, all of which is available in it's final format in introductory_neural_network.ipynb

This project is relatively minimal, but it requires Python 3 (Preferably as distributed by Anaconda), as well as a few additional packages that can be installed via:

    conda install numpy matplotlib pandas jupyter notebook

In this project, we implemented all the steps that go into a very simple neural network (pictured above), including an implementation of Iterative Gradient Descent using Activation Functions, Forward Propogation and Backward Propogation

The final model yielded a final (MSE) Training loss of 0.068 and Validation loss of 0.169 using 5000 epochs, a learning rate of 0.45, and 10 hidden nodes, which can be considered effecient performance.

100 Iterations	500 iterations	1000 Iterations

0.01 Learning Rate	0.15 Learning Rate	0.5 Learning Rate

10 Nodes	100 Nodes	250 Nodes