python - "unfair" Pandas 分类.from_codes

Question

我必须为分类数据分配标签。让我们考虑一下 iris 示例:

import pandas as pd
import numpy as np
from sklearn.datasets import load_iris

iris = load_iris()

print "targets: ", np.unique(iris.target)
print "targets: ", iris.target.shape
print "target_names: ", np.unique(iris.target_names)
print "target_names: ", iris.target_names.shape

将打印:

targets: [0 1 2] targets: (150L,) target_names: ['setosa' 'versicolor' 'virginica'] target_names: (3L,)

为了生成所需的标签，我使用 pandas.Categorical.from_codes:

print pd.Categorical.from_codes(iris.target, iris.target_names)

[setosa, setosa, setosa, setosa, setosa, ..., virginica, virginica, virginica, virginica, virginica] Length: 150 Categories (3, object): [setosa, versicolor, virginica]

让我们尝试一下不同的示例:

# I define new targets
target = np.array([123,123,54,123,123,54,2,54,2])
target = np.array([1,1,3,1,1,3,2,3,2])
target_names = np.array(['paglia','gioele','papa'])
#---
print "targets: ", np.unique(target)
print "targets: ", target.shape
print "target_names: ", np.unique(target_names)
print "target_names: ", target_names.shape

如果我再次尝试转换标签中的分类值:

print pd.Categorical.from_codes(target, target_names) 

我收到错误消息:

C:\Users\ianni\Anaconda2\lib\site-packages\pandas\core\categorical.pyc in from_codes(cls, codes, categories, ordered) 459 460 if len(codes) and (codes.max() >= len(categories) or codes.min() < -1): --> 461 raise ValueError("codes need to be between -1 and " 462 "len(categories)-1") 463

ValueError: codes need to be between -1 and len(categories)-1

你知道为什么吗？

Answer 1

Do you know why?

如果您仔细查看错误回溯:

In [128]: pd.Categorical.from_codes(target, target_names)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-128-c2b4f6ac2369> in <module>()
----> 1 pd.Categorical.from_codes(target, target_names)

~\Anaconda3_5.0\envs\py36\lib\site-packages\pandas\core\categorical.py in from_codes(cls, codes, categories, ordered)
    619
    620         if len(codes) and (codes.max() >= len(categories) or codes.min() < -1):
--> 621             raise ValueError("codes need to be between -1 and "
    622                              "len(categories)-1")
    623

ValueError: codes need to be between -1 and len(categories)-1

您将看到满足以下条件:

codes.max() >= len(categories)

就你的情况而言:

In [133]: target.max() >= len(target_names)
Out[133]: True

换句话说，pd.Categorical.from_codes() 期望 codes 为从 0 开始到 len(categories ) - 1

解决方法:

In [173]: target
Out[173]: array([123, 123,  54, 123, 123,  54,   2,  54,   2])

辅助字典:

In [174]: mapping = dict(zip(np.unique(target), np.arange(len(target_names))))

In [175]: mapping
Out[175]: {2: 0, 54: 1, 123: 2}

In [176]: reverse_mapping = {v:k for k,v in mapping.items()}

In [177]: reverse_mapping
Out[177]: {0: 2, 1: 54, 2: 123}

构建分类系列:

In [178]: ser = pd.Categorical.from_codes(pd.Series(target).map(mapping), target_names)

In [179]: ser
Out[179]:
[papa, papa, gioele, papa, papa, gioele, paglia, gioele, paglia]
Categories (3, object): [paglia, gioele, papa]

反向映射:

In [180]: pd.Series(ser.codes).map(reverse_mapping)
Out[180]:
0    123
1    123
2     54
3    123
4    123
5     54
6      2
7     54
8      2
dtype: int64