How can I perform optimisation on a trained machine learning regression model in Matlab?

Hello,

I have trained a Gaussian process regression model in Matlab. It is of the “struct” form trainedModel.predictFcn(x) where x is my input vector.

How can I use the function in this form for optimisation?

I essentially want to solve and find all values of the vector x for which the function equals some scalar value Y.

Thanks!

[Asvin Kumar]

a. Optimization on or Optimization using? What are you trying to optimize?

b. Where do you intend to use this trained model?

c. You mentioned, finding all values of x for which rainedModel.predictFcn(x) == y. What's your search space?

These details could help anyone answer this question better.

[Hamid Attar]

Hi Asvin,

Please see the figure below. The contour is showing part of the function trainedModel.predictFcn(x). This functon has 6 dimentions, since I trained the model with 6 feature vectors, but I just kept 4 variables constant to visulise this contour for the purpose of asking this question.

I want to solve the model for all non unique solutions where the model intercepts with a hyperplane set as a constant Y (20 for example).

In 2D this can be visulised in the figure below. I want to solve the function trainedModel.predictFcn(x) for all non unique solutions of x along the intersecting boundary, depicted by the red line Y. That is, I want to solve for all non unique variable combinations, x, in the 6D space, such that the values of trainedModel.predictFcn(x) are equal to Y at all of those non unique values of x.

I intend to use this trained model in Matlab only. I would just like to locase the variable combinations of this boundary (In the full 6D space).

[Asvin Kumar]

I'm going to let someone else from the community take this. This is a bit out of my field.