WHY Support Vector Regression (SVR) considers EXCLUDED ROWS during analysis?

[#1]

I am trying to predict continuous response on the basis of continuous inputs. However, during analysis, I've observed that SVR is considering excluded rows, which I want to use for model validation. Furthermore, the regression results couldn't be duplicated in Scikit-Learn library.

Please help on this issue!

[#2 · Accepted solution]

Hello,

I just ran a default SVM model on the same data set, Tablet Production.jmp

I'm using JMP16.2 on a Mac

I randomly selected 50% of the rows and set their state to Hidden/Excluded

I ran the SVM platform from the Analyze menu with and without the hidden/excluded state.

It looks like the platform is obeying the row state.

Why isn't SciKit Matching JMP? It likely has to do with how The model was parameterized.

For example, the default Fit might not match the default in SciKit.

Support Vector Machines(
	Y( :Dissolution ),
	X(
		:API Particle Size, :Mill Time, :Screen Size, :Mag. Stearate Supplier,
		:Lactose Supplier, :Sugar Supplier, :Talc Supplier, :Blend Time,
		:Blend Speed, :Compressor, :Force, :Coating Supplier, :Coating Viscosity,
		:Inlet Temp, :Exhaust Temp, :Spray Rate, :Atomizer Pressure
	),
	Fit(
		Kernel Function( "Radial Basis Function" ),
		Gamma( 0.0588235294117647 ),
		Cost( 1 ),
		Validation Method( "None" )
	)
);

[#3]

Thanks for your detailed response, I've also upgraded to 16.2, reanalysed the problem and now the issue is not duplicating (excluded rows are also not appearing in predicted/residual plots). I am also able to replicate the analysis on scikit-Learn by using gamma='auto' or (1/features). Thanks again for your detailed answer, obliged!