应用一种热编码后，数据和输入测试集的 n 个特征不同

Question

我一直在尝试训练 RandomForestRegressor，以根据给定的训练集预测给定测试集的房价。

import pandas as pd
import numpy as np
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_absolute_error
from sklearn.metrics import accuracy_score
from sklearn.preprocessing import LabelEncoder
from sklearn.preprocessing import OneHotEncoder
from sklearn.preprocessing import MaxAbsScaler

file='file:///F:/Download sort required/train.csv'
data=pd.read_csv(file)
data.dropna(axis=0,subset=['SalePrice'],inplace=True)
y=data.SalePrice

predictors=['LotArea','OverallQual','GrLivArea','GarageCars','TotRmsAbvGrd','Neighborhood','HouseStyle','YearBuilt','ExterQual','KitchenQual']

One_hot_encoded_predictors=['Neighborhood','HouseStyle','YearBuilt','ExterQual','KitchenQual']

X_uncoded=data[predictors]
#Encoding the training data
X_uncoded=pd.get_dummies(X_uncoded,columns=One_hot_encoded_predictors)
X=X_uncoded


maxabsscaler=MaxAbsScaler()
X_max_abs=maxabsscaler.fit_transform(X)
model=RandomForestRegressor()
model.fit(X_max_abs,y)

test_file='file:///C:/Users/shand/Downloads/test.csv'
test_data=pd.read_csv(test_file)
X_uncoded_test=test_data[predictors]
X_uncoded_test=pd.get_dummies(X_uncoded_test,columns=One_hot_encoded_predictors)
X_test=X_uncoded_test
X_test.fillna(X_test.mean(),inplace=True)
X_max_abs_test=maxabsscaler.fit_transform(X_test)

predicted_prices=model.predict(X_max_abs_test)

my_submission = pd.DataFrame({'Id': test_data.Id, 'SalePrice': predicted_prices})
my_submission.to_csv('submission.csv', index=False)

我对分类特征应用了一种热编码，然后是 maxabsscaler 转换，因为大多数数据在 -1 到 1 或 0 到 1 之间变化。但是编译时的代码会引发以下错误-

> >  28 X_test.fillna(X_test.mean(),inplace=True)
>      29 X_max_abs_test=maxabsscaler.fit_transform(X_test)
> ---> 30 predicted_prices=model.predict(X_max_abs_test)
>      31 my_submission = pd.DataFrame({'Id': test_data.Id, 'SalePrice': predicted_prices})
>      32 my_submission.to_csv('submission.csv', index=False)
> 
> C:\Users\shand\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py
> in predict(self, X)
>     683         """
>     684         # Check data
> --> 685         X = self._validate_X_predict(X)
>     686 
>     687         # Assign chunk of trees to jobs
> 
> C:\Users\shand\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py
> in _validate_X_predict(self, X)
>     353                                  "call `fit` before exploiting the model.")
>     354 
> --> 355         return self.estimators_[0]._validate_X_predict(X, check_input=True)
>     356 
>     357     @property
> 
> C:\Users\shand\Anaconda3\lib\site-packages\sklearn\tree\tree.py in
> _validate_X_predict(self, X, check_input)
>     374                              "match the input. Model n_features is %s and "
>     375                              "input n_features is %s "
> --> 376                              % (self.n_features_, n_features))
>     377 
>     378         return X
> 
> ValueError: Number of features of the model must match the input.
> Model n_features is 158 and input n_features is 151

在应用一个热编码和 maxabsscaler 后，有 158 个特征用于训练模型。 尽管我对训练集和测试集数据都应用了相同的转换，但谁能解释为什么我会收到此错误？ 我应该怎么做才能纠正这个错误？

PS 数据来自 - https://www.kaggle.com/c/house-prices-advanced-regression-techniques

Answer 1

正如您提到的，编码后训练数据和测试数据的列数不同。 训练数据有 158 列，而测试数据只有 151 列。

#Encoding the train data
X_uncoded=pd.get_dummies(X_uncoded,columns=One_hot_encoded_predictors)
X=X_uncoded
print(X.shape)
(1460, 158)
#Encoding the test data
X_uncoded_test=pd.get_dummies(X_uncoded_test,columns=One_hot_encoded_predictors)
print(X_uncoded_test.shape)
(1459, 151)

这可能是因为测试数据的级别数少于训练数据。请参阅下面来自pandas.get_dummies的示例

import pandas as pd
s = pd.Series(list('abca'))
pd.get_dummies(s)
   a  b  c
0  1  0  0
1  0  1  0
2  0  0  1
3  1  0  0

您可以考虑在编码之前将训练和测试结合起来，然后按照here 的描述在编码后将它们分离回训练和测试