NN for bike sharing prediction

The purpose of this project is to develop a Neural Network (NN) to predict bike sharing ride and is my first project of the Udacity Deep Learning Nanodegree program.

Quickstart

This project is based on Pytorch and Jupyter notebook. All the installation process are handled by Anaconda, one of the the most popular Data Science platform, so is quite straightforward.

Download and install the Anaconda platform for the OS and architecture in use:

$ wget https://repo.anaconda.com/archive/Anaconda3-2019.03-Linux-x86_64.sh
$ ./Anaconda3-2019.03-Linux-x86_64.sh

Create an Anaconda environment for the project based on Python 3.6:

$ conda install python=3.6.8
$ conda create -n dog-breed python=3.6
$ conda activate dog-breed

Install all the required dependencies always using Anaconda:

$ conda install -c conda-forge matplotlib
$ conda install numpy jupyter notebook
$ conda install pytorch-cpu torchvision-cpu -c pytorch

Dataset

The dataset is based on the Bike Sharing Dataset by Hadi Fanaee-T of the Laboratory of Artificial Intelligence and Decision Support (LIAAD), University of Porto.

Goal

The aim of the project is to design the Forward and Backward pass of the Neural Network, and tune the Hyper-parameters to achieve a training loss below 0.09 and a validation loss below 0.18.

Results

The project submitted shown a training loss of 0.063 and a validation loss of 0.139, and meet all the requirements:

• Code Functionality
  • All the code in the notebook runs in Python 3 without failing
  • All unit tests pass.
  • The sigmoid activation function is implemented correctly.
• Forward Pass
  • The forward pass is correctly implemented for the network's training.
  • The run method correctly produces the desired regression output for the neural network.
• Backward Pass
  • The network correctly implements the backward pass for each batch, correctly updating the weight change.
  • Updates to both the input-to-hidden and hidden-to-output weights are implemented correctly.
• Hyperparameters:
  • The number of epochs is chosen such the network is trained well enough to accurately make predictions but is not overfitting to the training data.
  • The number of hidden units is chosen such that the network is able to accurately predict the number of bike riders, is able to generalize, and is not overfitting.
  • The learning rate is chosen such that the network successfully converges, but is still time efficient.
  • The number of output nodes is properly selected to solve the desired problem.
  • The training loss is below 0.09 and the validation loss is below 0.18.