possible bug in _create_initial_point GeneralizedGammaRegressionFitter

[josarago]

I am trying to create a cure version of the GeneralizedGammaRegressionFitter and looking into the code. My understanding was that the _create_initial_point() method is used to fit a univariate model and use the parameters to initialize the intercepts of the regression model. Is that correct?
If so, I wonder if there could be a bug. If I follow the example in the doc:

from lifelines import GeneralizedGammaRegressionFitter
from lifelines.datasets import load_rossi

df = load_rossi()
df['Intercept'] = 1.

# this will regress df against all 3 parameters
ggf = GeneralizedGammaRegressionFitter(penalizer=1.).fit(df, 'week', 'arrest')
ggf.print_summary()

# If we want fine control over the parameters <-> covariates.
# The values in the dict become can be formulas, or column names in lists:
regressors = {
    'mu_': df.columns.difference(['arrest', 'week']),
    'sigma_': ["age", "Intercept"],
    'lambda_': 'age + 1',
}

ggf = GeneralizedGammaRegressionFitter(penalizer=0.0001).fit(df, 'week', 'arrest', regressors=regressors)
ggf.print_summary()

the model converges and fit the data but if I look into the value of constant_col, I see that (Xs.var(0) < 1e-8).idxmax()
is equal to ('sigma_', 'Intercept').
I was expecting something like:

[
  ('sigma_', 'Intercept'),
  ('mu_', 'Intercept'),
  ('lambda_', 'Intercept')
]

As a matter of fact, all the intercept columns verify: (Xs.var(0) < 1e-8) but I don't understand why we take the max index?

Is this intentional or are we missing mu_ and lambda_ intercepts?