使用具有 logloss 和 RFECV 的不平衡数据集的问题

Question

我在 RFECV 中使用不平衡数据集 (54:38:7%) 进行特征选择，如下所示：

# making a multi logloss metric
from sklearn.metrics import log_loss, make_scorer
log_loss_rfe = make_scorer(score_func=log_loss, greater_is_better=False)

# initiating Light GBM classifier
lgb_rfe = LGBMClassifier(objective='multiclass', learning_rate=0.01, verbose=0, force_col_wise=True,
                         random_state=100, n_estimators=5_000, n_jobs=7)

# initiating RFECV
rfe = RFECV(estimator=lgb_rfe, min_features_to_select=2, verbose=3, n_jobs=2, cv=3, scoring=log_loss_rfe)
# fitting it
rfe.fit(X=X_train, y=y_train)

我得到了一个错误，大概是因为 sklearn 的 RFECV 制作的子样本没有包含我数据中的所有类。在 RFECV 之外拟合相同的数据时我没有任何问题。

这是完整的错误：

---------------------------------------------------------------------------

_RemoteTraceback                          Traceback (most recent call last)

_RemoteTraceback: 
"""
Traceback (most recent call last):
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/joblib/externals/loky/process_executor.py", line 431, in _process_worker
    r = call_item()
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/joblib/externals/loky/process_executor.py", line 285, in __call__
    return self.fn(*self.args, **self.kwargs)
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/joblib/_parallel_backends.py", line 595, in __call__
    return self.func(*args, **kwargs)
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/joblib/parallel.py", line 262, in __call__
    return [func(*args, **kwargs)
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/joblib/parallel.py", line 262, in <listcomp>
    return [func(*args, **kwargs)
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/utils/fixes.py", line 222, in __call__
    return self.function(*args, **kwargs)
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/feature_selection/_rfe.py", line 37, in _rfe_single_fit
    return rfe._fit(
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/feature_selection/_rfe.py", line 259, in _fit
    self.scores_.append(step_score(estimator, features))
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/feature_selection/_rfe.py", line 39, in <lambda>
    lambda estimator, features: _score(
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/model_selection/_validation.py", line 674, in _score
    scores = scorer(estimator, X_test, y_test)
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/metrics/_scorer.py", line 199, in __call__
    return self._score(partial(_cached_call, None), estimator, X, y_true,
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/metrics/_scorer.py", line 242, in _score
    return self._sign * self._score_func(y_true, y_pred,
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/utils/validation.py", line 63, in inner_f
    return f(*args, **kwargs)
  File "/home/ubuntu/ds_jup_venv/lib/python3.8/site-packages/sklearn/metrics/_classification.py", line 2265, in log_loss
    raise ValueError("y_true and y_pred contain different number of "
ValueError: y_true and y_pred contain different number of classes 3, 2. Please provide the true labels explicitly through the labels argument. Classes found in y_true: [0 1 2]
"""


The above exception was the direct cause of the following exception:

ValueError                                Traceback (most recent call last)

<ipython-input-9-5feb62a6f457> in <module>
      1 rfe = RFECV(estimator=lgb_rfe, min_features_to_select=2, verbose=3, n_jobs=2, cv=3, scoring=log_loss_rfe)
----> 2 rfe.fit(X=X_train, y=y_train)

~/ds_jup_venv/lib/python3.8/site-packages/sklearn/feature_selection/_rfe.py in fit(self, X, y, groups)
    603             func = delayed(_rfe_single_fit)
    604 
--> 605         scores = parallel(
    606             func(rfe, self.estimator, X, y, train, test, scorer)
    607             for train, test in cv.split(X, y, groups))

~/ds_jup_venv/lib/python3.8/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

~/ds_jup_venv/lib/python3.8/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())

~/ds_jup_venv/lib/python3.8/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

1 frames

/usr/lib/python3.8/concurrent/futures/_base.py in __get_result(self)
    386     def __get_result(self):
    387         if self._exception:
--> 388             raise self._exception
    389         else:
    390             return self._result

ValueError: y_true and y_pred contain different number of classes 3, 2. Please provide the true labels explicitly through the labels argument. Classes found in y_true: [0 1 2]

如何解决这个问题以便能够递归地选择特征？

Answer 1

对数损失需要概率预测，而不是类别预测，所以你应该添加

log_loss_rfe = make_scorer(score_func=log_loss, needs_proba=True, greater_is_better=False)

错误是因为没有它，传递的y_pred 是一维的（类 0、1、2）和sklearn assumes it's a binary classification 问题，这些预测是正类的概率。为了解决这个问题，它增加了负类的概率，但是与你的三个类相比只有两列。

Answer 2

考虑应用分层交叉验证，它将尝试保留每个类的样本分数。试验以下 scikit-learn 交叉验证器之一： sklearn.model_selection.StratifiedKFold, StratifiedShuffleSplit, RepeatedStratifiedKFold，将 RFECV 中的 cv=3 替换为所选的交叉验证器。

编辑 我错过了StratifiedKFold 是RFECV 中的默认交叉验证器这一事实。实际上，该错误与log_loss_rfe 有关，它是用needs_proba=False 定义的。感谢@BenReiniger！