将标签转换回原始编码

Question

我有一张这样的桌子：

           exterior_color interior_color  ... isTheftRecovered    price
0            Night Black        Unknown  ...                0  16995.0
1    Orca Black Metallic        Unknown  ...                0  17995.0
2        Brilliant Black        Unknown  ...                0   9995.0
3  Azores Green Metallic   Nougat Brown  ...                0  24495.0
4                  Black          Brown  ...                0  16990.0

代码：

from sklearn.preprocessing import LabelEncoder
from sklearn import tree
import pandas as pd
import numpy as np


data_frame = pd.read_csv()


le = LabelEncoder()


cols = ['exterior_color', 'interior_color', 'location', 'make', 'model', 'mileage', 'style', 'year', 'engine', 'accidentCount',
        'accidentCount', 'ownerCount', 'isCleanTitle', 'isFrameDamaged', 'isLemon', 'isSalvage', 'isTheftRecovered', 'price']


data_frame[cols] = data_frame[cols].apply(LabelEncoder().fit_transform)


exclude_price = data_frame[data_frame.columns.difference(['price'])]


clf = tree.DecisionTreeClassifier()
clf = clf.fit(exclude_price, data_frame.price)

my_data = ['Night Black', 'Unknown', 'Patchogue, NY', 'Audi', 'Q7', '5000', 'S-line 3.0T quattro',
           '2015', '2.0L Inline-4 Gas Turbocharged', '0', '5.0', '1', '1', '0', '0', '1']

new_data = le.fit_transform(my_data)

answer = clf.predict([new_data])

print(f"Car's price has been predicted ${answer[0]}")

这段代码将对数据进行标记编码，然后预测给定数据的价格，但我无法将标签转换回原始编码，并使用inverse_transform来显示实际价格。

Answer 1

通过以apply(LabelEncoder().fit_transform)方式编码，您将失去对编码器对象的访问权限。相反，您可以将它们保存在按列名键的encoder字典中：

from collections import defaultdict
encoder = defaultdict(LabelEncoder)

df[cols] = df[cols].apply(lambda x: encoder[x.name].fit_transform(x))

然后通过encoder['price']解码最终价格

decoded = encoder['price'].inverse_transform(answer)[0]
print(f"Car's price has been predicted as ${decoded:.2f}")

# Car's price has been predicted as $16995.00