TemplateExpressionSpec don't use variables names and .sympy() doesn't work

[Vahan-tech]

Hello to everyone I am using TemplateExpressionSpec a follows but unfortunally the i,r and k are written in the results with #1, #2 and #3 as variables instead of the specified variable_names

template = TemplateExpressionSpec(
    expressions=["i", "r","k"],
    variable_names=['x1', 'x2', 'x3'],
    combine="i(x1,x2,x3)*(r(x2)/k(x3))",
)

Even specifying the variable names in .fit() doesn't help

model.fit(X, y, variable_names=["x1", "x2", "x3"])

Another problem is related to the fact that .sympy() returns that is not usable on template expression spec and if I use .get_best() i only obtain the single functions separated by commas (so not combined as specified) with #1,#2 and #3 as variables in the section lambda_format and equation.

Any clue on how to fix this ?
My goal would be to achieve the combined expression i(x1,x2,x3)*(r(x2)/k(x3)) with the specified variable names and to have the sympy and equation format

[MilesCranmer]

Hi @Vahan-tech,

This is correct and is the intended behavior. Template expressions print using #1 instead of "x1" because the specific arguments can change. For example your template could be f(x1) - f(x2). So the only consistent way to handle this is to print #1 to mean "first argument" rather than specific variable names.

Perhaps a workaround could be add by someone in the future.

.sympy() is not available with template expressions – this is intended. See #911 for the issue for this.

Cheers,
Miles