how to write the equation for expected prediction value of a gpr model in symbolic form

1 次查看（过去 30 天）
Vivek Annem
Vivek Annem 2019-11-4
I am trying to get the prediction function of fitgpr in symbolic form.
My model type is Linear basis, exact FitMethod and exact PredictMethod. I kept the kernel function as default so it is squared exponential.
In this article, it says if using exact gpr method, i can use this equation for prediction
I coded(attached below) up the function in symbolic form and when i substitute xe to 0 to check the output, my value is off. I checked with other values of xe as well and my symbolic equation output came out wrong at all the values tested.
What am I doing wrong?
Here is my code
clear all
clc
close all
rng(0,'twister'); % For reproducibility
n = 11; % n - size of dataset
c = 5; % c - drag co-efficient
xi = 0; % xi - lower bound
xf = 5; % xf - upper bound
var = 0.2; % v - variance
x = linspace(xi,xf,n)'; % populating x
y = (-c * x) + (var * randn(n,1)); % y - eqn
gprMdl = fitrgp(x,y,'Basis','Linear',...
'FitMethod','exact','PredictMethod','exact');
syms v f
sig_l = gprMdl.KernelInformation.KernelParameters(1);
sig_f = gprMdl.KernelInformation.KernelParameters(2);
f = [v 1] * [gprMdl.Beta];
for i=1:n
f = f + ((gprMdl.Alpha(i) * (sig_f^2)) * exp( -0.5 * (sig_l^(-2)) * ( v - x(i) ) ) ) ;
end
eval(subs(f,0))
predict(gprMdl,0)

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Gaussian Process Regression 的更多信息

标签

产品


版本

R2017b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by