Formulae does not respect original names in interactions and function calls

[tomicapretto]

It was a very bad choice to decide to format tokens according to my personal taste. The result is that formulae modifies the original name of the term. See x : y and f(x= z) below. Note the spaces around : and after =.

import pandas as pd
from formulae import design_matrices

df = pd.DataFrame(
    {
        "x": list("ab"),
        "y": list("cd"),
        "z": [1, 2]
    }
)

def f(x):
    return x

dm = design_matrices("0 + x : y + f(x= z)", df)
dm.common.terms

{'x:y': Term([Variable(x), Variable(y)]), 'f(x = z)': Term([Call(f(x = z))])}

But formulae says there are two terms, x:y without spaces and f(x = z) with spaces on both sides.

Let's see what other formula parsing libraries do.

Patsy

from patsy import dmatrix
dm = dmatrix("0 + x : y + f(x = z)", df)
dm.design_info.term_names

['x:y', 'f(x=z)']

Patsy also removes the space around : and formats the function call, removing spaces around the equal symbol.

Formulaic

from formulaic import model_matrix
mm = model_matrix("0 + x : y + f(x = z)", df)
mm.model_spec.feature_names

['f(x= z)', 'x[T.a]:y[T.c]', 'x[T.a]:y[T.d]', 'x[T.b]:y[T.c]', 'x[T.b]:y[T.d]']

Formulaic removes spaces around : but it keeps the function call as it is.

I think that Formulaic is right. Operators such as : can be formatted, but we should not touch function calls.