[BUG] TweedieGLM never passes alpha to TweedieRegressor #1584

Are you on the latest chainladder version?

• Yes, this bug occurs on the latest version (main at 0e5a7a9: chainladder 0.10.1, Python 3.12.15, numpy 2.3.3, pandas 2.3.3, scikit-learn 1.7.2).

Describe the bug in words

TweedieGLM stores alpha (glm.py L208) but builds its TweedieRegressor without it (L227–L235), so every fit uses scikit-learn's default penalty of 1.0. alpha=0 therefore doesn't give the unpenalized GLM the docstring describes (L64–L68): on ukmotor its IBNR comes out 4.2% above chain ladder. No commit in the history has passed alpha.

How can the bug be reproduced?

import chainladder as cl

ukmotor = cl.load_sample("ukmotor")
glm = cl.TweedieGLM(alpha=0).fit(ukmotor)
print(glm.model_.estimator_ml.named_steps["model"].alpha)
print(cl.Chainladder().fit(glm.transform(ukmotor)).ibnr_.sum().round(1))
print(cl.Chainladder().fit(ukmotor).ibnr_.sum().round(1))

1.0
29866.2
28655.8

What is the expected behavior?

import chainladder as cl

ukmotor = cl.load_sample("ukmotor")
glm = cl.TweedieGLM(alpha=0).fit(ukmotor)
assert glm.model_.estimator_ml.named_steps["model"].alpha == 0
ibnr = cl.Chainladder().fit(glm.transform(ukmotor)).ibnr_.sum()
assert round(ibnr, 1) == round(cl.Chainladder().fit(ukmotor).ibnr_.sum(), 1)

With alpha=self.alpha passed, both asserts pass and the repro prints 0, 28655.8 and 28655.8. I have that one-line change locally with a regression test in test_glm.py. The default stays 1.0, so fits that don't set alpha are unchanged.

Would you be willing to contribute this ticket?

• Yes, absolutely!
• Yes, but I would like some help.
• No.

(This issue was written with Árbol, an AI assistant, under my direction.)