Same Multiple-output coregionalized predictions when adding more correlated outputs.

[essefi-ahlem]

Hey all,

I am using GPCoregionalizedRegression from GPy library.

This the context :
In some industrial processes, we aim predicting the concentration of the a variable (y1) , which is expensive to measure (we do have some values only of it), using inexpensive and oversampled variables as a proxy y2,y3(who are always available) that helps in predicting y1.

When tried giving single output varible y1 with gaps in it the model did a great job in interpolating the gaps.

But as I added more correlated outputs y2, y3, y4 on top of the most important output (in my case y1), I was expecting the predictions will get more accurate as it tries to use correlation between the outputs to interpolate unknown samples in y1.

but surprisingly, I had slightly the same results.

Is that a normal behavior?