如何从 sklearn 管道转换器中提取特征名称？

Question

供参考：

Python 3.8.3
sklearn 1.0.2

我有一个 scikit-learn pipeline 可以为我格式化一些数据，如下所述：

我这样定义我的 pipeline：

# Pipeline 1
cat_selector = make_column_selector(dtype_include=object)
num_selector = make_column_selector(dtype_include=np.number)

cat_linear_processor = OneHotEncoder(handle_unknown="ignore", drop='first', sparse=False)
num_linear_processor = make_pipeline(SimpleImputer(strategy="median", add_indicator=True), MinMaxScaler(feature_range=(-1,1)))

linear_preprocessor = make_column_transformer( (num_linear_processor, num_selector), (cat_linear_processor, cat_selector) )

model_params ={'alpha': 0.0013879181970625643,
 'l1_ratio': 0.9634269882730605,
 'fit_intercept': True,
 'normalize': False,
 'max_iter': 245.69684524349375,
 'tol': 0.01855761485447601,
 'positive': False,
 'selection': 'random'}
model = ElasticNet(**model_params)

pipeline = make_pipeline(linear_preprocessor, model)

pipeline.steps 产量：

[('columntransformer',
  ColumnTransformer(transformers=[('pipeline',
                                   Pipeline(steps=[('simpleimputer',
                                                    SimpleImputer(add_indicator=True,
                                                                  strategy='median')),
                                                   ('minmaxscaler',
                                                    MinMaxScaler(feature_range=(-1,
                                                                                1)))]),
                                   <sklearn.compose._column_transformer.make_column_selector object at 0x0000029CA3231EE0>),
                                  ('onehotencoder',
                                   OneHotEncoder(handle_unknown='ignore',
                                                 sparse=False),
                                   <sklearn.compose._column_transformer.make_column_selector object at 0x0000029CA542F040>)])),
 ('elasticnet',
  ElasticNet(alpha=0.0013879181970625643, l1_ratio=0.9634269882730605,
             max_iter=245.69684524349375, normalize=False, selection='random',
             tol=0.01855761485447601))]

我想做的是检索 trained/tested 上的数据的特征名称。

我已尝试引用许多其他问题：

但是，这些解决方案都没有奏效。例如：

[i for i in v.get_feature_names() for k, v in pipeline.named_steps.items() if hasattr(v,'get_feature_names')]

产量：

----> 1 [i for i in v.get_feature_names() for k, v in pipeline.named_steps.items() if hasattr(v,'get_feature_names')]

NameError: name 'v' is not defined

我试过了：

pipeline[:-1].get_feature_names_out()

产量：

AttributeError: Estimator simpleimputer does not provide get_feature_names_out. Did you mean to call pipeline[:-1].get_feature_names_out()?

如何从当前管道编码后检索特征名称？

Answer 1

我想这个 post 可能有帮助：

也就是说，问题应该只是sklearn的版本。我在几个月前 posted 中引用的 PR 似乎刚刚合并，尽管从那以后还没有新版本。安装实际的 sklearn 开发版本，scikit-learn 1.1.dev0 应该可以解决问题（至少对我来说是这样）。

您可以这样安装 nightly builds：pip install --pre --extra-index https://pypi.anaconda.org/scipy-wheels-nightly/simple scikit-learn -U。

这是一个关于 toy 数据集的例子：

import pandas as pd
import numpy as np
from sklearn.preprocessing import OneHotEncoder, MinMaxScaler
from sklearn.impute import SimpleImputer
from sklearn.pipeline import make_pipeline
from sklearn.compose import make_column_selector, make_column_transformer
from sklearn.linear_model import ElasticNet

X = pd.DataFrame({'city': ['London', 'London', 'Paris', 'Sallisaw', ''],
              'title': ['His Last Bow', 'How Watson Learned the Trick', 'A Moveable Feast', 'The Grapes of Wrath', 'The Jungle'],
              'expert_rating': [5, 3, 4, 5, 3],
              'user_rating': [4, 5, 4, 2, 3]})

# Pipeline 1
cat_selector = make_column_selector(dtype_include=object)
num_selector = make_column_selector(dtype_include=np.number)

cat_linear_processor = OneHotEncoder(handle_unknown="ignore", drop='first', sparse=False)
num_linear_processor = make_pipeline(SimpleImputer(strategy="median", add_indicator=True), MinMaxScaler(feature_range=(-1,1)))

linear_preprocessor = make_column_transformer( (num_linear_processor, num_selector), (cat_linear_processor, cat_selector) )

model_params ={
    'alpha': 0.0013879181970625643,
    'l1_ratio': 0.9634269882730605,
    'fit_intercept': True,
    'normalize': False,
    'max_iter': 245,
    'tol': 0.01855761485447601,
    'positive': False,
    'selection': 'random'}
model = ElasticNet(**model_params)

pipeline = make_pipeline(linear_preprocessor, model)
pipeline.fit(X.iloc[:, :-1], X.iloc[:, -1])

pipeline[:-1].get_feature_names_out()

如何从 sklearn 管道转换器中提取特征名称？

How to extract feature names from sklearn pipeline transformers?

python

scikit-learn

pipeline