Updating scikit-learn: 'SVC' object has no attribute '_probA'?

Question

We updated to Python 3.8.2 and are getting an error with scikit-learn:

Traceback (most recent call last):
File "manage.py", line 16, in <module>
execute_from_command_line(sys.argv)
File "/home/ubuntu/myWebApp/.venv/lib/python3.8/site-packages/django/core/management/__init__.py", line 381, in execute_from_command_line
utility.execute()
File "/home/ubuntu/myWebApp/.venv/lib/python3.8/site-packages/django/core/management/__init__.py", line 375, in execute
self.fetch_command(subcommand).run_from_argv(self.argv)
File "/home/ubuntu/myWebApp/.venv/lib/python3.8/site-packages/django/core/management/base.py", line 316, in run_from_argv
self.execute(*args, **cmd_options)
File "/home/ubuntu/myWebApp/.venv/lib/python3.8/site-packages/django/core/management/base.py", line 353, in execute
output = self.handle(*args, **options)
File "/home/ubuntu/myWebApp/server_modules/rss_ml_score/management/commands/rssmlscore.py", line 22, in handle
run.build_and_predict(days=options['days'], rescore=options['rescore'])
File "/home/ubuntu/myWebApp/server_modules/rss_ml_score/utils/run.py", line 96, in build_and_predict
predict_all(filename)
File "/home/ubuntu/myWebApp/server_modules/rss_ml_score/models/predict_model.py", line 135, in predict_all
voting_predicted_hard, voting_predicted_soft = predict_from_multiple_estimator(fitted_estimators, X_predict_list,
File "/home/ubuntu/myWebApp/server_modules/rss_ml_score/models/train_model.py", line 66, in predict_from_multiple_estimator
pred_prob1 = np.asarray([clf.predict_proba(X)
File "/home/ubuntu/myWebApp/server_modules/rss_ml_score/models/train_model.py", line 66, in <listcomp>
pred_prob1 = np.asarray([clf.predict_proba(X)
File "/home/ubuntu/myWebApp/.venv/lib/python3.8/site-packages/sklearn/svm/_base.py", line 662, in _predict_proba
if self.probA_.size == 0 or self.probB_.size == 0:
File "/home/ubuntu/myWebApp/.venv/lib/python3.8/site-packages/sklearn/svm/_base.py", line 759, in probA_
return self._probA
AttributeError: 'SVC' object has no attribute '_probA'

Do I need to use another library in addition to sci-kit learn to access _probA?

Update in response to a comment:

The line of code that's throwing the error is:

pred_prob1 = np.asarray([clf.predict_proba(X)
                         for clf, X in zip(estimators, X_list)])

...that calls this line in _base.py:

def _predict_proba(self, X):
    X = self._validate_for_predict(X)
    if self.probA_.size == 0 or self.probB_.size == 0:

...which calls this line, also in _base.py:

@property
def probA_(self):
    return self._probA

...which throws the error:

AttributeError: 'SVC' object has no attribute '_probA'

All this has worked fine for many months, but is not currently working, even after updating to the latest scikit-learn.

Answer 1

It turned out that I had to stay with the same version of sci-kit that was used to train the models we currently have (scikit-learn==0.21.2). Later versions of scikit don't work with our existing code / models. If we want to upgrade scikit, we have to retrain our models with the new version of scikit.