カテゴリ変数

Pythonで学ぶデータクリーニング

Adel Nehme

Content Developer @DataCamp

どんなエラーが起こり得るか？

I) 値の不一致

不一致な表記: 'married', 'Maried', 'UNMARRIED', 'not married'..
前後の空白: 'married ', ' married '..

II) 過多なカテゴリを少数に集約

新しい区分を作成: 連続の世帯収入から 0-20K, 20-40K など
区分を再マップ: 世帯収入カテゴリを 'rich', 'poor' の2区分に

III) データ型が category であることの確認（第1章で学習）

値の一貫性

大文字小文字: 'married', 'Married', 'UNMARRIED', 'unmarried'..

# 結婚状況列を取得
marriage_status = demographics['marriage_status']
marriage_status.value_counts()

unmarried    352
married      268
MARRIED      204
UNMARRIED    176
dtype: int64

値の一貫性

# DataFrameで件数集計
marriage_status.groupby('marriage_status').count()

                 household_income  gender
marriage_status                          
MARRIED                       204     204
UNMARRIED                     176     176
married                       268     268
unmarried                     352     352

値の一貫性

# 大文字化

marriage_status['marriage_status'] = marriage_status['marriage_status'].str.upper()
marriage_status['marriage_status'].value_counts()

UNMARRIED    528
MARRIED      472

# 小文字化

marriage_status['marriage_status'] = marriage_status['marriage_status'].str.lower()
marriage_status['marriage_status'].value_counts()

unmarried    528
married      472

値の一貫性

前後の空白: 'married ', 'married', 'unmarried', ' unmarried'..

# 結婚状況列を取得
marriage_status = demographics['marriage_status']
marriage_status.value_counts()

 unmarried   352
unmarried    268
married      204
married      176
dtype: int64

値の一貫性

# 余分な空白を削除
demographics = demographics['marriage_status'].str.strip()
demographics['marriage_status'].value_counts()

unmarried    528
married      472

カテゴリへの集約

データからカテゴリを作成: income 列から income_group 列を作る。

# Using qcut()
import pandas as pd
group_names = ['0-200K', '200K-500K', '500K+']
demographics['income_group'] = pd.qcut(demographics['household_income'], q = 3, 
                                       labels = group_names)
# Print income_group column
demographics[['income_group', 'household_income']]

     category  household_income
0   200K-500K  189243
1       500K+  778533
..

カテゴリへの集約

データからカテゴリを作成: income 列から income_group 列を作る。

# Using cut() - create category ranges and names
ranges = [0,200000,500000,np.inf]
group_names = ['0-200K', '200K-500K', '500K+']
# Create income group column
demographics['income_group'] = pd.cut(demographics['household_income'], bins=ranges, 
                                      labels=group_names)
demographics[['income_group', 'household_income']]

     category  Income
0      0-200K  189243
1       500K+  778533

カテゴリへの集約

カテゴリをより少数へ集約: カテゴリ列の区分を減らす。

operating_system 列の値: 'Microsoft', 'MacOS', 'IOS', 'Android', 'Linux'

operating_system 列の目標: 'DesktopOS', 'MobileOS'

# Create mapping dictionary and replace
mapping = {'Microsoft':'DesktopOS', 'MacOS':'DesktopOS', 'Linux':'DesktopOS',
           'IOS':'MobileOS', 'Android':'MobileOS'}
devices['operating_system'] = devices['operating_system'].replace(mapping)
devices['operating_system'].unique()

array(['DesktopOS', 'MobileOS'], dtype=object)

Passons à la pratique !

Pythonで学ぶデータクリーニング