Can't clone LinearGAM estimator object for sklearn cross-validation

[vmgustavo]

Bug

Can't use LinearGAM with ScikitLearn cross_validate as it get's a RuntimeError for not being able to clone the model.

The only estimator out of all the GAM options that can't be cloned is LinearGAM

To Reproduce

• using cross_validate

from sklearn.model_selection import cross_validate

from pygam import LinearGAM, s
from pygam.datasets import toy_interaction

X, y = toy_interaction(return_X_y=True)

gam = LinearGAM(s(0, by=1))

cross_validate(gam, X, y, cv=5, scoring='neg_mean_absolute_error')

~/miniconda3/envs/sbid/lib/python3.8/site-packages/sklearn/model_selection/_validation.py in <genexpr>(.0)
    242     scores = parallel(
    243         delayed(_fit_and_score)(
--> 244             clone(estimator), X, y, scorers, train, test, verbose, None,
    245             fit_params, return_train_score=return_train_score,
    246             return_times=True, return_estimator=return_estimator,

~/miniconda3/envs/sbid/lib/python3.8/site-packages/sklearn/utils/validation.py in inner_f(*args, **kwargs)
     70                           FutureWarning)
     71         kwargs.update({k: arg for k, arg in zip(sig.parameters, args)})
---> 72         return f(**kwargs)
     73     return inner_f
     74 

~/miniconda3/envs/sbid/lib/python3.8/site-packages/sklearn/base.py in clone(estimator, safe)
     94         param2 = params_set[name]
     95         if param1 is not param2:
---> 96             raise RuntimeError('Cannot clone object %s, as the constructor '
     97                                'either does not set or modifies parameter %s' %
     98                                (estimator, name))

RuntimeError: Cannot clone object LinearGAM(callbacks=['deviance', 'diffs'], fit_intercept=True, 
   max_iter=100, scale=None, terms=s(0), tol=0.0001, verbose=False), as the constructor either does not set or modifies parameter callbacks

• If only LinearGAM is not able to be cloned

from sklearn.base import clone

from pygam import LinearGAM, GammaGAM, InvGaussGAM, LogisticGAM, PoissonGAM, ExpectileGAM

for estimator in [LinearGAM, GammaGAM, InvGaussGAM, LogisticGAM, PoissonGAM, ExpectileGAM]:
    try:
        clone(estimator())
    except Exception as e:
        print(estimator.__name__, e.__class__.__name__)

LinearGAM RuntimeError

RuntimeError: Cannot clone object LinearGAM(callbacks=['deviance', 'diffs'], fit_intercept=True, 
   max_iter=100, scale=None, terms='auto', tol=0.0001, verbose=False), as the constructor either does not set or modifies parameter callbacks

Environment

Python 3.8.5
pygam==0.8.0

[lucasmazim]

Same error here!

[thistleknot]

I am having this issue as well

[miguelfmc]

Working on this right now.

It seems like this has to do with a small bug in the __init__ method of LinearGAM in which the callbacks parameter is not getting passed when calling super().__init__().

I tried the examples shared by @vmgustavo and got no runtime errors:

Example 1

from sklearn.model_selection import cross_validate
from pygam import LinearGAM, s
from pygam.datasets import toy_interaction

X, y = toy_interaction(return_X_y=True)

gam = LinearGAM(s(0, by=1))

cross_validate(gam, X, y, cv=5, scoring='neg_mean_absolute_error')

Example 2

from sklearn.base import clone

from pygam import LinearGAM, GammaGAM, InvGaussGAM, LogisticGAM, PoissonGAM, ExpectileGAM

for estimator in [LinearGAM, GammaGAM, InvGaussGAM, LogisticGAM, PoissonGAM, ExpectileGAM]:
    try:
        clone(estimator())
    except Exception as e:
        print(estimator.__name__, e.__class__.__name__)

Environment

Python 3.6.13
scikit-learn 0.24.2

Issue seems to be fixed so I will submit PR shortly.

[miguelfmc]

Just noticed that there was an open PR addressing this as well as the mutable default callbacks values:

#267

[Cain-Iriving]

from sklearn.model_selection import cross_validate from pygam import LinearGAM, s from pygam.datasets import toy_interaction X, y = toy_interaction(return_X_y=True) gam = LinearGAM(s(0, by=1)) cross_validate(gam, X, y, cv=5, scoring='neg_mean_absolute_error')

It doesn't work.

Cannot clone object LinearGAM(callbacks=['deviance', 'diffs'], fit_intercept=True, max_iter=100, scale=None, terms=s(0), tol=0.0001, verbose=False), as the constructor either does not set or modifies parameter callbacks

[jgu115]

Hi, to confirm, has the issue been fixed in any new release? I still see the problem so to confirm if the issue is fixed in any release

[kos-tso]

+1 I can also confirm this. LinearGAM throws the error, while LogisticGAM for example does not. As far as I can tell the issue is still the bug described by @miguelfmc , which has not been fixed, at least in the master branch. Notice that in the following part of the code the callbacks parameter is not passed anywhere:

pyGAM/pygam/pygam.py

Lines 2448 to 2469 in a6c14e4

	def __init__(
	self,
	terms='auto',
	max_iter=100,
	tol=1e-4,
	scale=None,
	callbacks=['deviance', 'diffs'],
	fit_intercept=True,
	verbose=False,
	**kwargs,
	):
	self.scale = scale
	super(LinearGAM, self).__init__(
	terms=terms,
	distribution=NormalDist(scale=self.scale),
	link='identity',
	max_iter=max_iter,
	tol=tol,
	fit_intercept=fit_intercept,
	verbose=verbose,
	**kwargs,
	)

[dswah]

Thanks for documenting this problem and identifying the fix.
This issue have been fixed by #359