Accompanying paper: Gaussian Process Regression for Estimating EM Ducting Within the Marine Atmospheric Boundary Layer. Hilarie Sit and Christopher J. Earls.
Gaussian process regression (GPR) using sci-kit learn for predicting evaporation duct height from EM propagation factor data. For noise-contaminated test inputs, the ground truth posterior predictive distribution by using 1000 Monte-Carlo (MC) samples with equal mixing proportions. Inverse-variance weighted proportions with a specified number of samples (aug_num) to calculate predictions and uncertainty is available. Dataset and results are located in their respective folders.
Specify name of csv data file and the testing ratio. If noise is applied to the propagation factor measurement, you can use --MC tag to calculate the ground truth and specify the --aug_num tag to use inverse-variance weighted proportions. Automatically performs GPR naively on clean and noise-contaminated test inputs. To exclude either, include --clean or --noise tags.
python gpr.py --csv case1 --ratio 20 --aug_num (5, 10)The attached bash script loops over different testing ratios.
bash run_gpr.sh