This project takes a first look at the mechanics inside a neural network - namely, a series of linear algebra operations mixed with nonlinear activation function application. The forward pass and backpropagation steps are manually implemented in my_answers.py using standard numpy functions. The network implemented therein is used to train on, and subsequently predict, bike rental volume as a function of time. The pandas library is used to read in the data formatted as a CSV file.
The results are found in the Your_first_neural_network.ipynb notebook and duplicated here.
The training and validation losses are plotted as a function of iterations.
After training, predictions using an independent test dataset are plotted here, along with the truth.
