TemplateExpressionSpec with custom operator

[ProPLr00]

Hi there, I am a newbie in using pysr, but may I check if is there a way to use a custom operator (Indicator function here) within the TemplateExpressionSpec?
For example:

template = TemplateExpressionSpec(
    function_symbols = ["f"],
    combine="((; f), (x0)) -> Indicator(f(x0))"
)
PySRRegressor(
    ...
    unary_operators=[
        ...
        "Indicator(x::T) where {T} = x > T(0) ? T(1) : T(0)"
    ],
    extra_sympy_mappings={
        "Indicator": lambda x: sympy.Piecewise((1, x > 0), (0, True))
    },
   ...
)

[MilesCranmer]

It’s a bit more complex (see other discussion threads with gm89uk) but as a simpler solution I would just turn it into a multiplication, like

((; f), (x0,)) -> (f(x0) > 0*x0) * (1 + 0*x0)

[MilesCranmer]

This is by design, because the inputs to an expression within a TemplateExpressionSpec are completely arbitrary. You could pass x0^2 instead of x0, or even do something like f(g(x0)). So the only general printing method is simply say #1 which means “first input”, whatever that may be. Actually you can call f multiple times with different inputs too, so we can’t even parse the feature out for printing, because it could change.

[ProPLr00]

Owh, this means that I could still track the features being fed into the model through their sequence, right? Let's say I have 8 features:
f1, f2, ..., f8 after I set them into the TemplateExpressionSpec like this:

template = TemplateExpressionSpec(
    function_symbols = ["f", "g"],
    combine="""((; f, g), (x0, x1, x2, x3, x4, x5, x6, x7)) -> begin
    fx = f(x0, x1, x2, x3, x4, x5, x6, x7)
    gx = g(x0, x1, x2, x3, x4, x5, x6, x7)
    product_x = map(*, fx.x, gx.x)
    ValidVector(product_x, fx.valid && gx.valid)
    end """
)

Then during the output, the features will be parsed as follow:
f1 = x0 = #1, f2 = x1 = #2, ... , f8 = x7 = #8

If yes then thank you so much for your help Miles!

[MilesCranmer]

Yeah exactly

[MilesCranmer]

Btw you can just do

template = TemplateExpressionSpec(
    function_symbols = ["f", "g"],
    combine="""((; f, g), (x0, x1, x2, x3, x4, x5, x6, x7)) -> begin
    fx = f(x0, x1, x2, x3, x4, x5, x6, x7)
    gx = g(x0, x1, x2, x3, x4, x5, x6, x7)
    fx * gx
end """
)

The only reason you had to manually unpack it is because the Indicator function wasn’t defined on ValidVector. But * is, so you can just use that

[ProPLr00]

Nice and thanks Miles, really appreciate your help!