看起来你可以利用category类型的优势。

import pandas as pd

train = pd.DataFrame({'feature':['a', 'b', 'c', 'd']})
test = pd.DataFrame({'feature':['a']})

train['feature'] = train['feature'].astype('category')
dummies_type = train['feature'].dtype
test['feature'] = test['feature'].astype(dummies_type)

训练数据：

pd.get_dummies(train)

feature_a   feature_b   feature_c   feature_d
1   0   0   0
0   1   0   0
0   0   1   0
0   0   0   1

检测数据：

pd.get_dummies(test)

feature_a   feature_b   feature_c   feature_d
1   0   0   0

功能的新值：

test_oov = pd.DataFrame({'feature':['z']})
test_oov['feature'] = test_oov['feature'].astype(dummies_type)
pd.get_dummies(test_oov)

feature_a   feature_b   feature_c   feature_d
0   0   0   0

您可以从单个新观测创建虚拟值，然后使用原始指标矩阵中的列重新索引此帧列：

import pandas as pd
df = pd.DataFrame({'cat':['a','b','c','d'],'val':[1,2,5,10]})
df1 = pd.get_dummies(pd.DataFrame({'cat':['a'],'val':[1]}))
dummies_frame = pd.get_dummies(df)
df1.reindex(columns = dummies_frame.columns, fill_value=0)

退货：

val     cat_a   cat_b   cat_c   cat_d
  0     1       1       0       0       0

获取JAB的答案，以便在sklearn管道中使用它，此代码可能会帮助您：

from sklearn.base import BaseEstimator, TransformerMixin

class GetDummies(BaseEstimator, TransformerMixin):
    def __init__(self, dummy_columns):
        self.columns = None
        self.dummy_columns = dummy_columns

    def fit(self, X, y=None):
        self.columns = pd.get_dummies(X, columns=self.dummy_columns).columns
        return self

    def transform(self, X):
        X_new = pd.get_dummies(X, columns=self.dummy_columns)
        return X_new.reindex(columns=self.columns, fill_value=0)