Why does ndcg_score result in nan values? - python-3.x

Question

Consider the following code:
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import classification_report, ndcg_score, make_scorer
from sklearn.svm import SVC
X_data = pd.DataFrame(np.random.randint(0,1,size=(100, 4)), columns=list('ABCD'))
X_data = sp.csr_matrix(X_data.to_numpy())
Y_data = pd.DataFrame(np.random.choice([0,1,5], 100), columns=['Y'])
# Set the parameters by cross-validation
param_grid = {'kernel': ['rbf'], 'gamma': [1e-3, 1e-4],
'C': [1, 10, 100, 1000]}
clf = GridSearchCV(SVC(), param_grid, scoring=ndcg_score, refit=True, verbose=3, n_jobs=-1, error_score='raise')
test = clf.fit(X_data, Y_data)
I am wondering why this would raise the following error:
Fitting 5 folds for each of 8 candidates, totalling 40 fits
---------------------------------------------------------------------------
_RemoteTraceback Traceback (most recent call last)
_RemoteTraceback:
"""
Traceback (most recent call last):
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\joblib\externals\loky\process_executor.py", line 431, in _process_worker
r = call_item()
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\joblib\externals\loky\process_executor.py", line 285, in __call__
return self.fn(*self.args, **self.kwargs)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\joblib\_parallel_backends.py", line 595, in __call__
return self.func(*args, **kwargs)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\joblib\parallel.py", line 262, in __call__
return [func(*args, **kwargs)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\joblib\parallel.py", line 262, in <listcomp>
return [func(*args, **kwargs)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\utils\fixes.py", line 222, in __call__
return self.function(*args, **kwargs)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\model_selection\_validation.py", line 625, in _fit_and_score
test_scores = _score(estimator, X_test, y_test, scorer, error_score)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\model_selection\_validation.py", line 687, in _score
scores = scorer(estimator, X_test, y_test)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\utils\validation.py", line 74, in inner_f
return f(**kwargs)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\metrics\_ranking.py", line 1564, in ndcg_score
y_true = check_array(y_true, ensure_2d=False)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\utils\validation.py", line 63, in inner_f
return f(*args, **kwargs)
File "C:\Users\test\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\utils\validation.py", line 710, in check_array
array = array.astype(np.float64)
TypeError: float() argument must be a string or a number, not 'SVC'
"""
The above exception was the direct cause of the following exception:
TypeError Traceback (most recent call last)
<ipython-input-45-93a8890b095c> in <module>
18
19 clf = GridSearchCV(SVC(), param_grid, scoring=ndcg_score, refit=True, verbose=3, n_jobs=-1, error_score='raise')
---> 20 test = clf.fit(X_data, Y_data)
21 #print(test.best_score_)
~\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\utils\validation.py in inner_f(*args, **kwargs)
61 extra_args = len(args) - len(all_args)
62 if extra_args <= 0:
---> 63 return f(*args, **kwargs)
64
65 # extra_args > 0
~\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\model_selection\_search.py in fit(self, X, y, groups, **fit_params)
839 return results
840
--> 841 self._run_search(evaluate_candidates)
842
843 # multimetric is determined here because in the case of a callable
~\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\model_selection\_search.py in _run_search(self, evaluate_candidates)
1294 def _run_search(self, evaluate_candidates):
1295 """Search all candidates in param_grid"""
-> 1296 evaluate_candidates(ParameterGrid(self.param_grid))
1297
1298
~\Anaconda3\envs\kaggleSVM\lib\site-packages\sklearn\model_selection\_search.py in evaluate_candidates(candidate_params, cv, more_results)
793 n_splits, n_candidates, n_candidates * n_splits))
794
--> 795 out = parallel(delayed(_fit_and_score)(clone(base_estimator),
796 X, y,
797 train=train, test=test,
~\Anaconda3\envs\kaggleSVM\lib\site-packages\joblib\parallel.py in __call__(self, iterable)
1052
1053 with self._backend.retrieval_context():
-> 1054 self.retrieve()
1055 # Make sure that we get a last message telling us we are done
1056 elapsed_time = time.time() - self._start_time
~\Anaconda3\envs\kaggleSVM\lib\site-packages\joblib\parallel.py in retrieve(self)
931 try:
932 if getattr(self._backend, 'supports_timeout', False):
--> 933 self._output.extend(job.get(timeout=self.timeout))
934 else:
935 self._output.extend(job.get())
~\Anaconda3\envs\kaggleSVM\lib\site-packages\joblib\_parallel_backends.py in wrap_future_result(future, timeout)
540 AsyncResults.get from multiprocessing."""
541 try:
--> 542 return future.result(timeout=timeout)
543 except CfTimeoutError as e:
544 raise TimeoutError from e
~\Anaconda3\envs\kaggleSVM\lib\concurrent\futures\_base.py in result(self, timeout)
442 raise CancelledError()
443 elif self._state == FINISHED:
--> 444 return self.__get_result()
445 else:
446 raise TimeoutError()
~\Anaconda3\envs\kaggleSVM\lib\concurrent\futures\_base.py in __get_result(self)
387 if self._exception:
388 try:
--> 389 raise self._exception
390 finally:
391 # Break a reference cycle with the exception in self._exception
TypeError: float() argument must be a string or a number, not 'SVC'
I am not quite sure why this would result in a TypeError.

Answer 1

I cannot recreate the error you are reporting, but using error_score="raise" and n_jobs=1 (not strictly necessary, but the output is a little easier to read), and wrapping ndcg_score with make_scorer with needs_proba=True, I get this one:
Only ('multilabel-indicator', 'continuous-multioutput', 'multiclass-multioutput') formats are supported. Got multiclass instead
which supports my first comment: NDCG assumes multilabel format. That suggests you need to understand whether NDCG is really appropriate for your task, and if so either turn your problem into a multilabel one or write a custom scorer that converts the multiclass output into a multilabel (one-hot encoded) one before computing the score.