我最近了解了自定义转换器，并正在尝试将其实现为NLP管道的类别编码器。

转换器接收功能并按照下表所示的方式对其进行编码。

请帮助我以更Python化的方式编写此代码。我想我不应该在转换函数中使用y，如何解决这个问题？

#Custom Transformer that returns passed column as response coding and deletes the original column
from sklearn.base import BaseEstimator,TransformerMixin

class response_coding_transformer( BaseEstimator,TransformerMixin ):

  #Class Constructor
  def __init__( self,feature,y):
    self._feature = feature
    self._y = y

  #https://github.com/Lalit-Yadav-E1483/ml_case_studies/blob/7da02bdef5512c6ab636727a2a589072b0e2d515/cancer_diagnosis/initial_analysis.ipynb
  def response_table(alpha,X,y):
      value_count = X[feature].value_counts()
      gv_dict = dict()
      y_series = y[y.columns[0]]

      for i,denominator in value_count.items():
          vec = []
          for k in range(y_series.unique().size):
              cls_cnt = X.loc[(y_series==k) & (X[feature]==i)]
              vec.append((cls_cnt.shape[0] + alpha*10)/ (denominator + 90*alpha))
          gv_dict[i]=vec
      return gv_dict,y_series,value_count

  def response_coding(alpha,y):
      gv_dict,value_count = response_table(alpha,y)

      gv_fea = []
      for index,row in X.iterrows():
          if row[feature] in dict(value_count).keys():
              gv_fea.append(gv_dict[row[feature]])
          else:
              gv_fea.append([1/y_series.unique().size for _ in range(y_series.unique().size)])
      return np.array(gv_fea)

  def rc_df(response_array,feature):
    response_coded_df = pd.DataFrame()
    for i in range(response_array.shape[1]):
      name = feature + '_' + str(i)
      response_coded_df[name] = response_array[:,i]
    return response_coded_df

  #Return self nothing else to do here    
  def fit( self,y = None ):
    return self 

  #Method that describes what we need this transformer to do
  def transform( self,X):
    temp = pd.DataFrame()
    for feature in self._feature:
      alpha = 0
      response_array = response_coding(alpha,self._y)
      response_coded_df = rc_df(response_array,feature)
      temp = pd.concat([temp,response_coded_df],axis=1)
    new_x = X.drop(self._feature,axis = 1)
    return pd.concat([new_x,temp],axis=1)

这是管道的方法

#https://scikit-learn.org/stable/auto_examples/compose/plot_column_transformer_mixed_types.html
#https://stackoverflow.com/a/54704747/9292995

def Donors_choose_pipe(data,model,text_transformer):
  numeric_features = ['teacher_number_of_previously_posted_projects','price']
  numeric_transformer = Pipeline(steps=[
      ('scaler',MinmaxScaler())])

  categorical_features = ['school_state','teacher_prefix','project_grade_category','clean_categories','clean_subcategories']
  categorical_transformer = Pipeline(steps=[
      ('resp',response_coding_transformer(categorical_features))])

  # text_features = ['essay','project_title']
  text_transformer = Pipeline(steps=[
      ('trans',text_transformer)])

  preprocessor = ColumnTransformer(
      transformers=[('num',numeric_transformer,numeric_features),('cat',categorical_transformer,categorical_features),('essay',text_transformer,"essay"),# ('title',"project_title")
          ],n_jobs=-1,verbose=True,remainder = 'passthrough'
              )

  # Append classifier to preprocessing pipeline.
  # Now we have a full prediction pipeline.
  pipeline = Pipeline(steps=[('preprocessor',preprocessor),('classifier',model)])

  y = data['project_is_approved'].values
  X = data.drop(['project_is_approved'],axis=1)

  X_train,X_test,y_train,y_test = train_test_split(X,y,test_size=0.33,stratify=y)

  pipeline.fit(X_train,y_train)
  print("model score: %.3f" % pipeline.score(X_test,y_test))
  return pipeline,X_train,y_test


text_transformer = TfidfVectorizer(stop_words = 'english',min_df = 10)
model = XGBClassifier()
clf,y_test = Donors_choose_pipe(data,text_transformer)

这里是数据标题供参考

当我将y传递给转换器时，该代码似乎运行良好，但是我不知道如何在管道中进行操作。