Possible to set weight of features?

[JacobZhao]

I'm using SR to model a thermodynamical process.
The input "X" is a n-by-m matrix, where there are some feature in "m" is very important and should be include in the final equation.
In the basic SR process, like X(n_samples, n_features), does it possible to set different weight of features in X?
Or does it allow to force include specific feature?

[MilesCranmer]

You can write a custom loss function to penalize it if it doesn’t use the variable:

function my_loss(tree, dataset::Dataset{T,L}, options) where {T,L}
    prediction, flag = eval_tree_array(tree, dataset.X, options)
    if !flag
        return L(Inf)
    end
    y = dataset.y
    loss = L(sum(i -> abs2(prediction[i] - y[i]), eachindex(y)))

    # Check if any node in the tree is feature 5 
    # (note that Julia indexes from 1 rather than 0)
    contains_x5 = any(
        node -> (
            node.degree == 0
            && !(node.constant)
            && node.feature == 5
        ),
        tree
    )
    if !(contains_x5)
        # penalty term
        loss += L(1000)
    end
    return loss
end

Then you would pass this to the loss_function parameter (in PySR as a string; in Julia as the function name)

[JacobZhao]

Hi Miles, thx for your hint. Let me give a shot