使用 borutashap 时出现“无法从 'scipy.stats' 导入名称 'binom_test'”错误

Question

我正在尝试使用 borutashap 进行机器学习项目的特征选择。 我跑不动了：

from BorutaShap import BorutaShap

总是遇到这个错误：

ImportError: cannot import name 'binom_test' from 'scipy.stats'

我升级了 borutashap 和 scipy 以确保它们是最新版本。 我正在使用 conda env，我删除了整个 env 并重新创建了一个新的 env 并重新安装了所有库，但仍然是同样的错误。 有人可以帮我解决这个问题吗？预先感谢您。

Answer 1

函数

scipy.stats.binom_test()

已从 SciPy 1.12.0 中删除。 来源。

还有一个替代方案，

scipy.stats.binomtest()

，其名称略有不同。这是old和new的文档。

我相信您可以通过将 SciPy 降级到 1.11.0 或对 BorutaShap 进行修复以调用新的二项式测试函数来解决此问题。

我相信您需要以可编辑的形式安装软件包，然后编辑

Boruta-Shap/src/BorutaShap.py

来完成此操作，共两行：

9号线：

改变

from scipy.stats import binom_test, ks_2samp

到

from scipy.stats import binomtest, ks_2samp

885号线：

改变

        return [binom_test(x, n=n, p=p, alternative=alternative) for x in array]

到

        return [binomtest(x, n=n, p=p, alternative=alternative).pvalue for x in array]

另请参阅 BorutaShap 问题跟踪器上的此问题。

Answer 2

根据您的修订，我还在 'binomtest(x, n=n, p=p, Alternative=alternative).pvalue' 中遇到了另一个问题：

  0%|          | 0/100 [00:08<?, ?it/s]
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[75], line 10
      5 X_encoded = X#OrdinalEncoderPandas().fit_transform(X=X)
      6 Feature_Selector9 = BorutaShap(
      7     model=model, importance_measure="shap", classification=True
      8 )
---> 10 Feature_Selector9.fit(X=X_encoded, y=y, n_trials=100, random_state=0)
     12 # Returns Boxplot of features
     13 Feature_Selector9.plot(X_size=12, figsize=(8, 6), y_scale="log", which_features="all")

File /environment/miniconda3/envs/p39/lib/python3.9/site-packages/BorutaShap.py:473, in BorutaShap.fit(self, X, y, sample_weight, n_trials, random_state, sample, train_or_test, normalize, verbose, stratify)
    471         self.hits += hits
    472         self.history_hits = np.vstack((self.history_hits, self.hits))
--> 473         self.test_features(iteration=trial+1)
    475 self.store_feature_importance()
    476 self.calculate_rejected_accepted_tentative(verbose=verbose)

File /environment/miniconda3/envs/p39/lib/python3.9/site-packages/BorutaShap.py:934, in BorutaShap.test_features(self, iteration)
    919 def test_features(self, iteration):
    921     """
    922     For each feature with an undetermined importance perform a two-sided test of equality
    923     with the maximum shadow value to determine if it is statistcally better
   (...)
    931         Two arrays of the names of the accepted and rejected columns at that instance
    932     """
--> 934     acceptance_p_values = self.binomial_H0_test(self.hits,
    935                                                 n=iteration,
    936                                                 p=0.5,
    937                                                 alternative='greater')
    939     regect_p_values = self.binomial_H0_test(self.hits,
    940                                             n=iteration,
    941                                             p=0.5,
    942                                             alternative='less')
    944     # [1] as function returns a tuple

File /environment/miniconda3/envs/p39/lib/python3.9/site-packages/BorutaShap.py:885, in BorutaShap.binomial_H0_test(array, n, p, alternative)
    878 @staticmethod
    879 def binomial_H0_test(array, n, p, alternative):
    880     """
    881     Perform a test that the probability of success is p.
    882     This is an exact, two-sided test of the null hypothesis
    883     that the probability of success in a Bernoulli experiment is p
    884     """
--> 885     return [binomtest(x, n=n, p=p, alternative=alternative).pvalue for x in array]

File /environment/miniconda3/envs/p39/lib/python3.9/site-packages/BorutaShap.py:885, in <listcomp>(.0)
    878 @staticmethod
    879 def binomial_H0_test(array, n, p, alternative):
    880     """
    881     Perform a test that the probability of success is p.
    882     This is an exact, two-sided test of the null hypothesis
    883     that the probability of success in a Bernoulli experiment is p
    884     """
--> 885     return [binomtest(x, n=n, p=p, alternative=alternative).pvalue for x in array]

File /environment/miniconda3/envs/p39/lib/python3.9/site-packages/scipy/stats/_binomtest.py:287, in binomtest(k, n, p, alternative)
    202 def binomtest(k, n, p=0.5, alternative='two-sided'):
    203     """
    204     Perform a test that the probability of success is p.
    205 
   (...)
    285 
    286     """
--> 287     k = _validate_int(k, 'k', minimum=0)
    288     n = _validate_int(n, 'n', minimum=1)
    289     if k > n:

File /environment/miniconda3/envs/p39/lib/python3.9/site-packages/scipy/_lib/_util.py:352, in _validate_int(k, name, minimum)
    350     k = operator.index(k)
    351 except TypeError:
--> 352     raise TypeError(f'{name} must be an integer.') from None
    353 if minimum is not None and k < minimum:
    354     raise ValueError(f'{name} must be an integer not less '
    355                      f'than {minimum}') from None

TypeError: k must be an integer.

您能给一些建议吗？