Convert/decompose formula terms similarly to LinearModel

I'm trying to do something which must be implicitly done inside LinearModel.predict(), but I can't seem to get to it.

The task I'm trying to do is to have a linear formula given as a set of terms:

formulaTerms = {'X1','X2','X1*X2'}  % ie, y = X1 + X2 + X1*X2

and then have the input terms defined in their simplest state such as:

inputTerms = {'X1','X2'}

and then produce a function that will take in each term in inputTerms and return to me each term in formulaTerms. For example in this instance the result would be:

outFcn = @(input)[input(1), input(2), input(1)*input(2)]
outFcn([2 3]) % Which gets the 3 terms in the formula: [2 3 6]

What I'm trying to do is generalise this to a function such as:

function linearFcn = convert(formulaTerms, inputTerms)
    % linearFcn = ... (well, this is what I'm trying to write!)
end

The closest I can get to a general solution is something like the following where I can convert my formulaTerms into a formula string similar to what LinearModel.fit() accepts:

f = classreg.regr.LinearFormula('y ~ X1 + X2 + X1*X2',{'y','X1','X2'},'',[],'identity')

This is basically what is created inside the LinearModel object, but I don't know how LinearModel.predict(input) combines this input variable with the LinearFormula object to produce the actual terms that are summed up to complete the linear model prediction.

Answer 1

[Shashank Prasanna]

Sven, what you are interested in is the designmatrix: x2fx

Alternatively, if you are still interested in the what predict does:

mdl = LinearModel.fit(X,y)

formulaTerms=classreg.regr.modelutils.designmatrix([X ones(length(X),1)],...
'Model',mdl.Formula.Terms);

[Sven]

Aha... nice one Shashank, I'd found a designmatrix being built inside the LinearFunction object, but x2fx is indeed more suitable (and accessible). Thanks.