Deep Learning for Time Series Forecasting

A collection of examples for using DNNs for time series forecasting with Keras. The examples include:

• 0_data_setup.ipynb - set up data that are needed for the experiments
• 1_CNN_dilated.ipynb - dilated convolutional neural network model that predicts one step ahead with univariate time series
• 2_RNN.ipynb - recurrent neural network model that predicts one step ahead with univariate time series
• 3_RNN_encoder_decoder.ipynb - a simple recurrent neural network encoder-decoder approach to multi-step forecasting
• 4_ES_RNN.ipynb - a simplified exponential smoothing recurrent neural network model that predicts one step ahead with univariate time series

... and a number of hands-on exercises and demos.

Data

The data in all examples is from the GEFCom2014 energy forecasting competition¹. It consists of 3 years of hourly electricity load data from the New England ISO and also includes hourly temperature data. Before running the notebooks, download the data from https://www.dropbox.com/s/pqenrr2mcvl0hk9/GEFCom2014.zip?dl=0 and save it in the data folder.

¹Tao Hong, Pierre Pinson, Shu Fan, Hamidreza Zareipour, Alberto Troccoli and Rob J. Hyndman, "Probabilistic energy forecasting: Global Energy Forecasting Competition 2014 and beyond", International Journal of Forecasting, vol.32, no.3, pp 896-913, July-September, 2016.

Prerequisites

• Know the basics of Python
• Know the basics of neural networks
• Understand the basic concepts of machine learning and have some experience in building machine learning models
• Go through the following setup instructions to run notebooks on Azure Notebooks environment

Note: If you want to run the notebook in other environment, please check 'Requirements.txt' for a list of packages that you need to install.

Azure Notebooks Setup (1 Minute)

Microsoft Azure Notebooks is a free service that provides Jupyter Notebooks in cloud and has a support for R, Python and F#. We will use Microsoft Azure Notebook for this tutorial. Here are quick 3 steps to set it up:

1. Go to Azure Notebook page https://notebooks.azure.com/ and click 'Sign In' on the top right.
2. Use your Microsoft account to sign in. If you don't have a personal Microsoft account, you can click 'Create one' with any email address you have for free. (Note: You can use your personal Microsoft account. If you use your organizational account, you will need to go through the login process by your organization.)
3. If this is the first time you use Azure Notebook, you will need to create a user ID and click 'Save'. Now you are all set!

Tutorial Code and Data Setup (5 - 10 Minutes)

The following steps will guide you to setup code and data in your Azure Notebook environment for the tutorial.

1. Once you are logged into Azure Notebooks, go to 'My Projects' on the top left, and then click 'Upload GitHub Repo'.

2. In the pop out window, for 'GitHub repository' type in: 'Azure/DeepLearningForTimeSeriesForecasting'. Select 'Clone recursively'. Then type in any name you prefer for 'Project Name' and 'Project ID'. Once you have filled all boxes, click 'Import'. Please wait till you see a list of files cloned from git repository to your project.

3. Open the notebook '0_data_setup.ipynb'. Make sure you see 'Python 3.6' kernel on the top right. If not, you can select 'Kernel', then 'Change kernel' to make changes.

4. Run each cell in the notebook by click 'Run' on top. If you prefer to run all the cells together, click 'Cell' and select 'Run All'. This notebook will download a sample dataset to your environment and visualize the data. Please wait and make sure you can see all the visualizations.

Now you are all set! (Note: If you see errors return from the first code cell, it is very likely that the environment preparation is not finished yet. Please wait for 2 minutes and then go to 'Kernel', choose 'Restart and Clear Output' and rerun the cells.)