Supplementary Code for: Forecasting Urban Household Water Demand with Statistical and Machine Learning Methods Using Large Space-Time Data: A Comparative Study
This repository provides code and simulated data for training the models in the 2017 EMS paper "Forecasting Urban Household Water Demand with Statistical and Machine Learning Methods Using Large Space-Time Data: A Comparative Study."
Getting Started
The following instructions will get you a copy of the project up and running on your local machine. You can then change simulation settings or use the models on your own data.
Prerequisites
You will need:
- R, with packages
forecast,R.utils,caret,fields,mgcv,Rcpp, andRcppArmadilloinstalled - RTools (for Windows; other systems may need other compilers, e.g. here)
Installing
Once the prerequisites above have been installed, you simply need to copy the files in this repo to your local machine.
Running the code
| File | Description |
|---|---|
GAM-model-example-data.R |
Simulates data and fits the GAM model |
machine-learning-example-data.R |
Simulates data and fits the ML models |
ST-model-example-data.R |
Simulates data and fits the ST model |
time-series-example-data.R |
Simulates data and fits the time series models |
ST-model.cpp |
C++ code for the ST model |
NOIS.R |
Function to compute the NOIS validation metric |
gini.R |
Function to compute the GINI validation metric |
params.RData |
Contains parameters specific to ML methods script |
The C++ code ST-model.cpp provides the functions necessary for training the space-time model and using the model to make forecasts. The R script ST-model-example-data.R simulates some example data and fits and validates the ST model. The R script time-series-example-data.R simulates some example data and fits and validates the time series models. The R scripts GAM-model-example-data.R and machine-learning-example-data.R simulate some example data and fit and validate the corresponding model (either the GAM or the tree-based machine learning methods described in the paper).
This code should be able to be run as-is, either line-by-line or all at once. For example, calling source("ST-model-example-data.R") will simulate data and fit and validate the ST model on the simulated data, and show some traceplots of the MCMC chains as well as observed vs. predicted values from the ST model on a left-out data set.
Authors
- Hunter R. Merrill - Personal Site
- Isaac Duerr
- Chuan Wang
License
This project is licensed under the MIT License - see the LICENSE.md file for details