Fix trying to get pointer to None in svm/linear.pyx

[yosider]

This PR fixes the following bug in svm/linear.pyx:

After fitting LinearSVC and saving it using pickle, attempting to load and run predict() (or decision_function() and predict_proba() that call __sync_model()) results in the following error:

AttributeError                            Traceback (most recent call last)
----> 1 lin_svc_load.predict(X)

File .../lib/python3.10/site-packages/cuml/internals/api_decorators.py:188, in _make_decorator_function.<locals>.decorator_function.<locals>.decorator_closure.<locals>.wrapper(*args, **kwargs)
    185     set_api_output_dtype(output_dtype)
    187 if process_return:
--> 188     ret = func(*args, **kwargs)
    189 else:
    190     return func(*args, **kwargs)

File linear.pyx:693, in cuml.svm.linear.LinearSVM.predict()

File linear.pyx:667, in cuml.svm.linear.LinearSVM.__sync_model()

File linear.pyx:303, in cuml.svm.linear.LinearSVMWrapper.__cinit__()

AttributeError: 'NoneType' object has no attribute 'ptr'

I identified that the issue is caused by attempting to obtain a pointer to a variable with a None value in LinearSVMWrapper.__cinit__(). After fixing this and rebuilding, the code worked as expected.

Minimal code to reproduce the bug:

import pickle
import numpy as np
from cuml.svm import LinearSVC

X = np.random.randn(10, 2)
y = np.random.randint(0, 2, 10)

svc = LinearSVC()
svc.fit(X, y)

with open("tmp.pkl", "wb") as f:
    pickle.dump(svc, f)
with open("tmp.pkl", "rb") as f:
    svc_load = pickle.load(f)

svc_load.predict(X)

If this is not the appropriate way to fix the issue, I would appreciate any corrections.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[csadorf]

/ok to test

[csadorf]

@yosider Thanks a lot for reporting the issue, providing an MRE and even a proposed fix!

Before we move forward, I think it would be very important that we add a test that captures this bug. I would proposed to add a unit test to the python/cuml/tests/test_pickle.py module for the LinearSVC class similar to some of the other tests. Please let me know if you need more guidance!

[yosider]

@csadorf Thanks for your advice!
I've added test_linear_svc_pickle() based on test_svc_pickle().
It seems that LinearSVC does not support sparse data, so I've excluded tests for sparse cases.

[csadorf]

/ok to test

[csadorf]

LGTM! Thanks a lot!

[csadorf]

@yosider I'll look into the CI test failures which are almost certainly unrelated.

[dantegd]

/ok to test

[dantegd]

/merge