管道上的滚动平均值以训练模型

如何解决管道上的滚动平均值以训练模型

我在使用管道拟合模型时遇到问题，该管道希望添加具有某些特征的滚动平均值的列，然后训练模型。

数据框：

columns=['yr','mnth','hr','season','holiday','weekday','workingday','weathersit','temp','atemp','hum','windspeed','y']
df=pd.DataFrame(np.array([ [0,1,6,0.24,2.879,0.81,16],[0,0.22,2.727,0.80,40],2,32],3,0.75,13],4,1]]),columns=columns)

X_train=df.drop('y')
y_train=df['y']

对某些特征的滚动平均函数：

def rollingAv(Data):
          
    a=Data['atemp']
    a_shifted = a.shift(1)
    a_window = a_shifted.rolling(window=4)
    a_means = a_window.mean()
    Data['a_means'] = a_means

    h=Data['hum']
    h_shifted = h.shift(1)
    h_window = h_shifted.rolling(window=4)
    h_means = h_window.mean()
    Data['h_means'] = h_means

    w=Data['windspeed']
    w_shifted = w.shift(1)
    w_window = w_shifted.rolling(window=4)
    w_means = w_window.mean()
    Data['w_means'] = w_means

    Data=Data.dropna(subset=['a_means','h_means','w_means'])
    return Data.values

滚动平均类以适应和转换管道

class BikeRentalFeatureExtractor(BaseEstimator):
  
  def __init__(self):
    pass

  def fit(self,X,y=None):
    X=X.values
    if y.shape[0]>0:
      y=y[4:]
      return y
    else:
      pass
  
  def transform(x):
    return rollingAv(x)

管道和模型

model = Pipeline(steps=[
    ("extractor",BikeRentalFeatureExtractor()),("regressor",RandomForestRegressor())
    ])

parameters = {'regressor__n_estimators':[50,100,200,300]}

st = ShuffleSplit(n_splits=5,test_size=0.2,random_state=0)

clf = GridSearchCV(estimator=model,param_grid=parameters)

clf.fit(X_train,y_train)

我在 clf.fit(X_train,y_train) 之前没有错误，因为它似乎与数据相关，因为尽管我有以下消息，但我删除了该列，我再次尝试，问题在下一列继续：

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-21-86937c1966f0> in <module>()
----> 1 clf.fit(X_train,y_train)

12 frames
/usr/local/lib/python3.7/dist-packages/pandas/core/base.py in _try_aggregate_string_function(self,arg,*args,**kwargs)
    276 
    277         raise AttributeError(
--> 278             f"'{arg}' is not a valid function for '{type(self).__name__}' object"
    279         )
    280 

AttributeError: 'yr' is not a valid function for 'Series' object

解决方法

1. fit 假定返回 self
2. transform 是一个方法，应该将 self 作为第一个参数。