Sebastian_Ulrich
Reputation: 33
Using GroupShuffleSplit with GridSearchCV and cross_val_score for nested cross-validation
I am trying to perform nested cross-validation while also incorporating group-based splitting using the GroupShuffleSplit class. However, I'm encountering a "TypeError: cannot pickle 'generator' object" when trying to use a custom cross-validation object with GridSearchCV. As fas as i know this Error occurs because group_split.split(...) returns an generator which cant be used in the cross_val_score function. Therefore i want to ask if there is a way to easily use GroupShuffleSplit for nested cross-validation.
Regarding my simplified sample code:
I have a dataset with features X, labels y, and group labels groups. The goal is to perform nested cross-validation, where both the inner and outer loops split the data based on the group labels. I would like to use GridSearchCV for hyperparameter tuning and cross_val_score for evaluating the performance.
import numpy as np
from sklearn.model_selection import GroupShuffleSplit, GridSearchCV, cross_val_score
from sklearn.ensemble import RandomForestClassifier
X = np.random.rand(100, 10)
y = np.random.randint(2, size=100)
groups = np.random.randint(4, size=100) # Example group labels
rf_classifier = RandomForestClassifier()
param_grid = {'n_estimators': [50, 100, 200]}
inner_cv = GroupShuffleSplit(n_splits=5, test_size=0.2)
outer_cv = GroupShuffleSplit(n_splits=5, test_size=0.2)
grid_search = GridSearchCV(estimator=rf_classifier, param_grid=param_grid, cv=inner_cv.split(X, y, groups=groups))
nested_scores = cross_val_score(estimator=grid_search, X=X, y=y, cv=outer_cv.split(X, y, groups=groups))
Resulting in the following Stacktrace Error:
---------------------------------------------------------------------------
Empty Traceback (most recent call last)
File c:\Anaconda3_x64\lib\site-packages\joblib\parallel.py:825, in Parallel.dispatch_one_batch(self, iterator)
824 try:
--> 825 tasks = self._ready_batches.get(block=False)
826 except queue.Empty:
827 # slice the iterator n_jobs * batchsize items at a time. If the
828 # slice returns less than that, then the current batchsize puts
(...)
831 # accordingly to distribute evenly the last items between all
832 # workers.
File c:\Anaconda3_x64\lib\queue.py:168, in Queue.get(self, block, timeout)
167 if not self._qsize():
--> 168 raise Empty
169 elif timeout is None:
Empty:
During handling of the above exception, another exception occurred:
TypeError Traceback (most recent call last)
Cell In[29], line 16
13 outer_cv = GroupShuffleSplit(n_splits=5, test_size=0.2)
15 grid_search = GridSearchCV(estimator=rf_classifier, param_grid=param_grid, cv=inner_cv.split(X, y, groups=groups))
---> 16 nested_scores = cross_val_score(estimator=grid_search, X=X, y=y, cv=outer_cv.split(X, y, groups=groups))
18 print(nested_scores)
File c:\Anaconda3_x64\lib\site-packages\sklearn\model_selection\_validation.py:515, in cross_val_score(estimator, X, y, groups, scoring, cv, n_jobs, verbose, fit_params, pre_dispatch, error_score)
512 # To ensure multimetric format is not supported
513 scorer = check_scoring(estimator, scoring=scoring)
--> 515 cv_results = cross_validate(
516 estimator=estimator,
517 X=X,
518 y=y,
519 groups=groups,
520 scoring={"score": scorer},
521 cv=cv,
522 n_jobs=n_jobs,
523 verbose=verbose,
524 fit_params=fit_params,
525 pre_dispatch=pre_dispatch,
526 error_score=error_score,
527 )
528 return cv_results["test_score"]
File c:\Anaconda3_x64\lib\site-packages\sklearn\model_selection\_validation.py:266, in cross_validate(estimator, X, y, groups, scoring, cv, n_jobs, verbose, fit_params, pre_dispatch, return_train_score, return_estimator, error_score)
263 # We clone the estimator to make sure that all the folds are
264 # independent, and that it is pickle-able.
265 parallel = Parallel(n_jobs=n_jobs, verbose=verbose, pre_dispatch=pre_dispatch)
--> 266 results = parallel(
267 delayed(_fit_and_score)(
268 clone(estimator),
269 X,
270 y,
271 scorers,
272 train,
273 test,
274 verbose,
275 None,
276 fit_params,
277 return_train_score=return_train_score,
278 return_times=True,
279 return_estimator=return_estimator,
280 error_score=error_score,
281 )
282 for train, test in cv.split(X, y, groups)
283 )
285 _warn_or_raise_about_fit_failures(results, error_score)
287 # For callabe scoring, the return type is only know after calling. If the
288 # return type is a dictionary, the error scores can now be inserted with
289 # the correct key.
File c:\Anaconda3_x64\lib\site-packages\sklearn\utils\parallel.py:63, in Parallel.__call__(self, iterable)
58 config = get_config()
59 iterable_with_config = (
60 (_with_config(delayed_func, config), args, kwargs)
61 for delayed_func, args, kwargs in iterable
62 )
---> 63 return super().__call__(iterable_with_config)
File c:\Anaconda3_x64\lib\site-packages\joblib\parallel.py:1048, in Parallel.__call__(self, iterable)
1039 try:
1040 # Only set self._iterating to True if at least a batch
1041 # was dispatched. In particular this covers the edge
(...)
1045 # was very quick and its callback already dispatched all the
1046 # remaining jobs.
1047 self._iterating = False
-> 1048 if self.dispatch_one_batch(iterator):
1049 self._iterating = self._original_iterator is not None
1051 while self.dispatch_one_batch(iterator):
File c:\Anaconda3_x64\lib\site-packages\joblib\parallel.py:836, in Parallel.dispatch_one_batch(self, iterator)
833 n_jobs = self._cached_effective_n_jobs
834 big_batch_size = batch_size * n_jobs
--> 836 islice = list(itertools.islice(iterator, big_batch_size))
837 if len(islice) == 0:
838 return False
File c:\Anaconda3_x64\lib\site-packages\sklearn\utils\parallel.py:59, in <genexpr>(.0)
54 # Capture the thread-local scikit-learn configuration at the time
55 # Parallel.__call__ is issued since the tasks can be dispatched
56 # in a different thread depending on the backend and on the value of
57 # pre_dispatch and n_jobs.
58 config = get_config()
---> 59 iterable_with_config = (
60 (_with_config(delayed_func, config), args, kwargs)
61 for delayed_func, args, kwargs in iterable
62 )
63 return super().__call__(iterable_with_config)
File c:\Anaconda3_x64\lib\site-packages\sklearn\model_selection\_validation.py:268, in <genexpr>(.0)
263 # We clone the estimator to make sure that all the folds are
264 # independent, and that it is pickle-able.
265 parallel = Parallel(n_jobs=n_jobs, verbose=verbose, pre_dispatch=pre_dispatch)
266 results = parallel(
267 delayed(_fit_and_score)(
--> 268 clone(estimator),
269 X,
270 y,
271 scorers,
272 train,
273 test,
274 verbose,
275 None,
276 fit_params,
277 return_train_score=return_train_score,
278 return_times=True,
279 return_estimator=return_estimator,
280 error_score=error_score,
281 )
282 for train, test in cv.split(X, y, groups)
283 )
285 _warn_or_raise_about_fit_failures(results, error_score)
287 # For callabe scoring, the return type is only know after calling. If the
288 # return type is a dictionary, the error scores can now be inserted with
289 # the correct key.
File c:\Anaconda3_x64\lib\site-packages\sklearn\base.py:89, in clone(estimator, safe)
87 new_object_params = estimator.get_params(deep=False)
88 for name, param in new_object_params.items():
---> 89 new_object_params[name] = clone(param, safe=False)
90 new_object = klass(**new_object_params)
91 params_set = new_object.get_params(deep=False)
File c:\Anaconda3_x64\lib\site-packages\sklearn\base.py:70, in clone(estimator, safe)
68 elif not hasattr(estimator, "get_params") or isinstance(estimator, type):
69 if not safe:
---> 70 return copy.deepcopy(estimator)
71 else:
72 if isinstance(estimator, type):
File c:\Anaconda3_x64\lib\copy.py:161, in deepcopy(x, memo, _nil)
159 reductor = getattr(x, "__reduce_ex__", None)
160 if reductor is not None:
--> 161 rv = reductor(4)
162 else:
163 reductor = getattr(x, "__reduce__", None)
TypeError: cannot pickle 'generator' object
Upvotes: 1
Views: 191
Answers (1)
Ben Reiniger
Reputation: 12698
I'm not sure that this is possible before version 1.3 without writing a manual loop to replace cross_val_score. Besides the generator issue, you're trying to tell the grid search object that it should split all of X, but it won't see all of X (it having already been split by the outer splitter).
In 1.3, we get metadata routing which automatically routes groups to group splitters. Then we can do e.g.
from sklearn import set_config
set_config(enable_metadata_routing=True)
grid_search = GridSearchCV(estimator=rf_classifier, param_grid=param_grid, cv=inner_cv)
nested_scores = cross_val_score(estimator=grid_search, X=X, y=y, cv=outer_cv, params={'groups': groups})
Just to check that this really routes to both splitters, here's a modified version of your script:
import numpy as np
import pandas as pd
from sklearn.model_selection import GroupShuffleSplit, GridSearchCV, cross_val_score
from sklearn.base import BaseEstimator, ClassifierMixin
from sklearn import set_config
set_config(enable_metadata_routing=True)
X = np.random.rand(100, 10)
y = np.random.randint(2, size=100)
groups = np.random.randint(4, size=100) # Example group labels
X = pd.DataFrame(X)
class MyClassifier(ClassifierMixin, BaseEstimator):
def __init__(self, n_estimators=1):
self.n_estimators = n_estimators
def fit(self, X, y):
print("train: ", groups[X.index])
return self
def predict(self, X):
print("test: ", groups[X.index])
return np.random.randint(2, size=len(X))
rf_classifier = MyClassifier()
param_grid = {'n_estimators': [50, 100]}
inner_cv = GroupShuffleSplit(n_splits=2, test_size=0.33)
outer_cv = GroupShuffleSplit(n_splits=2, test_size=0.25)
grid_search = GridSearchCV(estimator=rf_classifier, param_grid=param_grid, cv=inner_cv, verbose=10)
nested_scores = cross_val_score(estimator=grid_search, X=X, y=y, cv=outer_cv, params={'groups': groups}, verbose=10)
print(nested_score)
The outer splits put a single group in the test set, then the inner splits pick one of the remaining three as test and the last two in train. Here's my output:
[CV] START .....................................................................
Fitting 2 folds for each of 2 candidates, totalling 4 fits
[CV 1/2; 1/2] START n_estimators=50.............................................
train: [1 3 3 3 1 3 3 1 1 1 1 1 3 1 1 1 3 3 1 3 3 3 3 1 1 1 3 3 3 3 3 3 3 3 3 1 3
3 3 3 1 3 1 1 1 3 3 1 1 3 1 1 1 1 1 1]
test: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[CV 1/2; 1/2] END ..............n_estimators=50;, score=0.353 total time= 0.0s
[CV 2/2; 1/2] START n_estimators=50.............................................
train: [3 3 3 3 3 0 0 3 3 3 3 3 3 0 3 0 0 0 3 3 0 0 0 3 3 0 0 3 3 3 3 3 3 3 3 3 0
0 0 0 3 3 3 0 0 3]
test: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]
[CV 2/2; 1/2] END ..............n_estimators=50;, score=0.407 total time= 0.0s
[CV 1/2; 2/2] START n_estimators=100............................................
train: [1 3 3 3 1 3 3 1 1 1 1 1 3 1 1 1 3 3 1 3 3 3 3 1 1 1 3 3 3 3 3 3 3 3 3 1 3
3 3 3 1 3 1 1 1 3 3 1 1 3 1 1 1 1 1 1]
test: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[CV 1/2; 2/2] END .............n_estimators=100;, score=0.412 total time= 0.0s
[CV 2/2; 2/2] START n_estimators=100............................................
train: [3 3 3 3 3 0 0 3 3 3 3 3 3 0 3 0 0 0 3 3 0 0 0 3 3 0 0 3 3 3 3 3 3 3 3 3 0
0 0 0 3 3 3 0 0 3]
test: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]
[CV 2/2; 2/2] END .............n_estimators=100;, score=0.333 total time= 0.0s
train: [1 3 3 3 1 3 3 1 1 1 0 1 1 0 3 1 1 1 3 3 1 3 3 3 0 3 1 1 0 1 0 0 3 3 0 0 0
3 3 0 0 3 3 3 3 3 1 3 3 3 3 0 0 1 0 0 3 1 1 1 3 3 1 1 0 0 3 1 1 1 1 1 1]
test: [2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2]
[CV] END ................................ score: (test=0.481) total time= 0.0s
[CV] START .....................................................................
Fitting 2 folds for each of 2 candidates, totalling 4 fits
[CV 1/2; 1/2] START n_estimators=50.............................................
train: [3 3 3 3 2 3 2 2 2 3 2 3 2 3 2 2 2 3 3 2 3 3 2 2 2 3 3 2 2 3 3 2 2 2 3 3 3
3 3 3 3 2 3 3 2 2 3 2 2 2 2 3 3 2 3 2]
test: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[CV 1/2; 1/2] END ..............n_estimators=50;, score=0.588 total time= 0.0s
[CV 2/2; 1/2] START n_estimators=50.............................................
train: [3 3 3 3 2 3 2 2 2 3 2 3 2 3 2 2 2 3 3 2 3 3 2 2 2 3 3 2 2 3 3 2 2 2 3 3 3
3 3 3 3 2 3 3 2 2 3 2 2 2 2 3 3 2 3 2]
test: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[CV 2/2; 1/2] END ..............n_estimators=50;, score=0.588 total time= 0.0s
[CV 1/2; 2/2] START n_estimators=100............................................
train: [3 3 3 3 2 3 2 2 2 3 2 3 2 3 2 2 2 3 3 2 3 3 2 2 2 3 3 2 2 3 3 2 2 2 3 3 3
3 3 3 3 2 3 3 2 2 3 2 2 2 2 3 3 2 3 2]
test: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[CV 1/2; 2/2] END .............n_estimators=100;, score=0.647 total time= 0.0s
[CV 2/2; 2/2] START n_estimators=100............................................
train: [3 3 3 3 2 3 2 2 2 3 2 3 2 3 2 2 2 3 3 2 3 3 2 2 2 3 3 2 2 3 3 2 2 2 3 3 3
3 3 3 3 2 3 3 2 2 3 2 2 2 2 3 3 2 3 2]
test: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[CV 2/2; 2/2] END .............n_estimators=100;, score=0.471 total time= 0.0s
train: [3 3 3 3 2 3 2 2 0 2 0 3 2 3 2 3 2 2 2 3 3 2 3 0 3 2 2 2 0 0 0 3 3 2 0 0 2
0 3 3 0 0 2 2 2 3 3 3 3 3 3 3 2 3 3 2 2 0 0 0 0 3 2 2 2 2 3 3 2 0 0 3 2]
test: [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]
[CV] END ................................ score: (test=0.593) total time= 0.0s
[0.48148148 0.59259259]
[Parallel(n_jobs=1)]: Done 1 tasks | elapsed: 0.0s
Upvotes: 1
Related Questions
- Iterating over dictionaries using 'for' loops
- ZeroDivisionError when using sklearn's BaggingClassifier with GridSearchCV
- K-Means clustering Hyperparameter Tuning
- GridSearchCV on a working pipeline returns ValueError
- What does ** (double star/asterisk) and * (star/asterisk) do for parameters?
- Random string generation with upper case letters and digits
- K-Means GridSearchCV hyperparameter tuning
- Error using GridSearchCV but not without GridSearchCV - Python 3.6.7
- Custom Sklearn Transformer works alone, Throws Error When Used in Pipeline