Restrict ypred in fitrgp function

I have a dataset where x is the independent variable and y is the dependent variable. I'm using the fitrgp function to model the response to the variable x given the values of y. I need to enforce fitrgp's ypred output to be between two values 'a' and 'b'.

How can I do?

Are there any options for fitrgp to enforce this constraint?

Thank you.

Answer 1

[Manas]

Hi Giacomo,

There is no direct way to enforce constraints on the predicted values. You may use the following code to enforce the constraints.

ypred(ypred < a) = a;  % Set values below 'a' to 'a'
ypred(ypred > b) = b;  % Set values above 'b' to 'b'

[Giacomo Caporusso]

Thanks for the reply. I hope that over time it will be possible to set the domain and codomain of the fitrgp function.

Answer 2

[Cyrus Monteiro]

The `fitrgp` function in MATLAB does not have a built-in option to enforce constraints on the predicted output (`ypred`) to be within a specific range. However, you can manually enforce this constraint after obtaining the predictions.

Here's an example of how you can enforce the constraint on `ypred` to be between two values, 'a' and 'b'

% Assuming you have already trained the Gaussian Process model using fitrgp
model = fitrgp(X, y);
% Obtain the predicted values
ypred = predict(model, X);
% Enforce the constraint on ypred to be between 'a' and 'b'
a = 0; % Lower bound
b = 1; % Upper bound
ypred_constrained = max(min(ypred, b), a);

In the code above, `ypred_constrained` is the predicted output (`ypred`) after enforcing the constraint to be between 'a' and 'b'. The `max` function limits the values of `ypred` to be less than or equal to 'b', and the `min` function ensures the values are greater than or equal to 'a'.

By applying this constraint manually, you can ensure that the predicted values (`ypred_constrained`) fall within the desired range.

More about the fitrgp here

[Giacomo Caporusso]

Thanks for the reply. I hope that over time it will be possible to set the domain and codomain of the fitrgp function.