是否可以为 OneHotEncoder 中的某些列指定 handle_unknown = 'ignore' 并为其他列指定 'error'？

Question

我有一个包含所有分类列的数据框，我使用来自sklearn.preprocessing 的oneHotEncoder 对其进行编码。我的代码如下：

from sklearn.preprocessing import OneHotEncoder
from sklearn.pipeline import Pipeline


steps = [('OneHotEncoder', OneHotEncoder(handle_unknown ='ignore')) ,('LReg', LinearRegression())]

pipeline = Pipeline(steps)

正如在OneHotEncoder 中看到的，handle_unknown 参数采用error 或ignore。我想知道是否有办法选择性地忽略某些列的未知类别而对其他列给出错误？

import pandas as pd

df = pd.DataFrame({'Country':['USA','USA','IND','UK','UK','UK'],
                   'Fruits':['Apple','Strawberry','Mango','Berries','Banana','Grape'],
                   'Flower':   ['Rose','Lily','Orchid','Petunia','Lotus','Dandelion'],
                   'Result':[1,2,3,4,5,6,]})

from sklearn.preprocessing import OneHotEncoder
from sklearn.linear_model import LinearRegression
from sklearn.pipeline import Pipeline

steps = [('OneHotEncoder', OneHotEncoder(handle_unknown ='ignore')) ,('LReg', LinearRegression())]

pipeline = Pipeline(steps)

from sklearn.model_selection import train_test_split

X = df[["Country","Flower","Fruits"]]
Y = df["Result"]
X_train, X_test, y_train, y_test = train_test_split(X,Y,test_size=0.3, random_state=30, shuffle =True)

print("X_train.shape:", X_train.shape)
print("y_train.shape:", y_train.shape)
print("X_test.shape:", X_test.shape)
print("y_test.shape:", y_test.shape)

pipeline.fit(X_train,y_train)

y_pred = pipeline.predict(X_test)

from sklearn.metrics import mean_squared_error
from sklearn.metrics import r2_score

#Mean Squared Error:
MSE = mean_squared_error(y_test,y_pred)

print("MSE", MSE)

#Root Mean Squared Error:
from math import sqrt

RMSE = sqrt(MSE)
print("RMSE", RMSE)

#R-squared score:
R2_score = r2_score(y_test,y_pred)

print("R2_score", R2_score)

在这种情况下，对于 Country、Fruits 和 Flowers 的所有列，如果有新值出现，模型仍然能够预测输出。

我想知道是否有办法忽略 Fruits 和 Flowers 的未知类别，但在 Country 列中引发未知值错误？

Answer 1

从 v0.20 开始，您可以使用 ColumnTransformer API。但是，对于旧版本，您可以轻松推出自己的预处理器实现，以选择性地处理列。

Here's a simple prototype I've implemented which extends OneHotEncoder. 您需要指定列列表以在raise_error_cols 参数上引发错误。任何未指定给此参数的列都被隐式处理为“忽略”。

样本运行

# Setup data
X_train

  Country     Flower  Fruits
2     IND     Orchid   Mango
0     USA       Rose   Apple
4      UK      Lotus  Banana
5      UK  Dandelion   Grape

X_test

  Country   Flower      Fruits
3      UK  Petunia     Berries
1     USA     Lily  Strawberry

X_test2 = X_test.append(
    {'Country': 'SA', 'Flower': 'Rose', 'Fruits': 'Tomato'}, ignore_index=True)
X_test2

  Country   Flower      Fruits
0      UK  Petunia     Berries
1     USA     Lily  Strawberry
2      SA     Rose      Tomato

from selective_handler_ohe import SelectiveHandlerOHE

she = SelectiveHandlerOHE(raise_error_cols=['Country'])
she.fit(X_train)

she.transform(X_test).toarray()
# array([[0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
#        [0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0.]])


she.transform(X_test2)
# ---------------------------------------------------------------------------
# ValueError: Found unknown categories SA in column Country during fit

Answer 2

我认为ColumnTransformer() 可以帮助您解决问题。您可以指定列表 您要为其应用 OneHotEncoder 的列，ignore 用于 handle_unknown，同样适用于 error。

使用ColumnTransformer将您的管道转换为以下内容

from sklearn.compose import ColumnTransformer

ct = ColumnTransformer([("ohe_ignore", OneHotEncoder(handle_unknown ='ignore'), 
                              ["Flower", "Fruits"]),
                        ("ohe_raise_error",  OneHotEncoder(handle_unknown ='error'),
                               ["Country"])])

steps = [('OneHotEncoder', ct),
         ('LReg', LinearRegression())]

pipeline = Pipeline(steps)

现在，当我们想要预测时

>>> pipeline.predict(pd.DataFrame({'Country': ['UK'], 'Fruits': ['Apple'], 'Flower': ['Rose']}))

array([2.83333333])

>>> pipeline.predict(pd.DataFrame({'Country': ['UK'], 'Fruits': ['chk'], 'Flower': ['Rose']}))

array([3.66666667])


>>> pipeline.predict(pd.DataFrame({'Country': ['chk'], 'Fruits': ['Apple'], 'Flower': ['Rose']}))

> ValueError: Found unknown categories ['chk'] in column 0 during
> transform

注意：ColumnTransformer 可从版本0.20 获得。