FastGP

Code for the paper: Fast Gaussian Process Approximations for Autocorrelated Data

Description:

This code repository contains the code to reproduce the tables and figures in the paper: Fast Gaussian Process Approximations for Autocorrelated Data. This document explains the data and provides instruction to execute the code.

Computer and software environment:

All experiments were implemented in R and executed on the Georgia Tech PACE high-performance computing cluster (RHEL7 Phoenix). The computations used Intel CPU-only nodes with 5 cores per job.

Code Execution:

Each of the tables and figures have their own scripts. Before running any script, make sure source.R is executed so all functions are loaded. The required packages are listed on top of this script.

Which results to reproduce	Data File	Code File	Output	Runtime
Figure 4	robot_data	figure_4.R	pacf_plot	3 seconds
Tables 1,2	robot_data	robot_table_1_2.R	table1_disp table2_disp	103.2 hours
Tables 3,4,5	LaGP website	robot_table_3_4_5.R	table3_disp table4_disp table5_disp	9.5 hours
Tables 6,7,9	DSWE book	robot_table_6_7_9.R	table6_disp table7_disp table9_disp	73.8 hours
Tables 8,9	DSWE book	robot_table_8_9.R	table8_disp table9_disp	23.6 hours
Figure 5	DSWE book	figure_5.R	T_num_plot	17.8 minutes
Figure 6	DSWE book	figure_6.R	stability_plot	3.2 hours

Note:

• Table 9 includes runtime for both dataset 5 and 6 of wind data in two rows. First row is created by wind5_table_6_7_9.R and second with wind6_table_8_9.R.
• Using a laptop to run the code may slightly slow down the approximation methods in general, but tempGP in particular can become prohibitively slow.

Datasets:

Three datasets are used for conducting numerical experiments, simulated robot arm data is uploaded to this package. The other datasets like laGP and DSWE data needs to be downloaded from their websites.

Script	Data
Figure 4 and Tables 1,2	These two scripts do use the robot_data from this package.
Tables 3,4,5	This script requires data from the laGP website. Download all files from laGP data and specify the directory path at the top of the script.
Tables 6,7,9	This script requires data from the DSWE book's dataset. Download all files from Dataset 5 and specify the directory path at the top of the script.
Tables 8,9 and Figure 5 and Figure 6	This script requires data from the DSWE book's dataset. Download all files from Dataset 6 and specify the directory path at the top of the script.

---