Errors when copying or pickling models using the expectation_propagation.EP() class

[olamarre]

Problem

When I try to deep copy or pickle a binary classification model implementing the EP() inference method (from the GPy.inference.latent_function_inference.expectation_propagation module), I get the following error:

  File "[...]/site-packages/GPy/inference/latent_function_inference/expectation_propagation.py", line 238, in __getstate__
    return [super(EPBase, self).__getstate__() , [self.epsilon, self.eta, self.delta]]
AttributeError: 'super' object has no attribute '__getstate__'

Reproducible example

Ran with Python 3.9 and GPy version 1.13.1:

import GPy
import copy
import numpy as np
import pickle as pkl

# Dummy binary classification dataset
X = np.array([0, 1, 2, 3]).reshape(-1, 1)
Y = np.array([0, 0, 1, 1]).reshape(-1, 1)

# Model
m = GPy.core.GP(
    X=X,
    Y=Y, 
    kernel=GPy.kern.RBF(input_dim=1, variance=1.0, lengthscale=1.0), 
    inference_method = GPy.inference.latent_function_inference.expectation_propagation.EP(),
    likelihood=GPy.likelihoods.Bernoulli(),
    mean_function=None
)

copy.deepcopy(m)  # triggers error
pkl.dumps(m)  # also triggers same error

Suggested fix

I believe the issue is that the parent class EPBase only inherits the basic object class. Calling super(EPBase, self).__getstate__() will fail because the object class does not implement this method. In fact, some other method calls with super() in the EPBase class should also fail (such as super(EPBase, self).__setstate__(state[0]) and super(EPBase, self)._save_to_input_dict()).

As a fix, I think these super() method calls can simply be removed in the EPBase class.

[MartinBubel]

Hi @olamarre

sorry for the late reply.
Thank you for the issue. Your suggested fix sounds promissing!
Could you help us by making a PR?

Best,
Martin

[olamarre]

No worries! Yes, I'll put together a PR.