A simple neural network with backpropagation that predicts daily bike rental ridership. 🚲 🚲
This is Project 1 of Udacity's Deep Learning Nanodegree Foundation program. The aim of this project is to go through the mathematical steps that a simple neural network with 1 hidden layer will take in order to learn patterns in the data.
The main steps include implementing a forward pass through the network, and calculating the output error and implementing the backpropogation algorithm. Then, I set the three hyperparameters for the network: the number of epochs, the learning rate, and the number of hidden nodes. The aim is to set the hyperparamters in such a way so that the training loss is low, but the network is not overfit to the training data. Next, the neural network is trained and validated over portions of the dataset, and the training and validation loss is visualized by a plot. Finally, the network is tested on a test set, and once again visualized to check if it is predicting accurately.
Jupyter Notebook and HTML.