テキスト列の結合とクリーニング

テキスト列を連結する

import pandas as pd

df = pd.DataFrame({
    'first_name': ['Alice', 'Bob', 'Carol'],
    'last_name': ['Smith', 'Jones', 'White']
})

df['full_name'] = df['first_name'] + ' ' + df['last_name']
print(df['full_name'].tolist())
# ['Alice Smith', 'Bob Jones', 'Carol White']

文字列以外の列と連結する

import pandas as pd

df = pd.DataFrame({
    'product': ['Widget', 'Gadget'],
    'version': [3, 5]
})

# Must convert int to str before concatenating
df['label'] = df['product'] + ' v' + df['version'].astype(str)
print(df['label'].tolist())
# ['Widget v3', 'Gadget v5']

.str.cat()で区切り文字を使って結合する

import pandas as pd
import numpy as np

df = pd.DataFrame({
    'city': ['NYC', 'LA', None],
    'state': ['NY', None, 'TX'],
    'country': ['USA', 'USA', 'USA']
})

# Join with ', ' — NaN rows produce NaN by default
df['location'] = df['city'].str.cat(
    [df['state'], df['country']],
    sep=', ',
    na_rep='N/A'
)
print(df['location'].tolist())
# ['NYC, NY, USA', 'LA, N/A, USA', 'N/A, TX, USA']

一貫性のないカテゴリラベルを正規化する

import pandas as pd

df = pd.DataFrame({
    'country': ['US', 'USA', 'United States', 'uk', 'UK', 'United Kingdom']
})

canonical = {
    'US': 'United States', 'USA': 'United States', 'United States': 'United States',
    'uk': 'United Kingdom', 'UK': 'United Kingdom', 'United Kingdom': 'United Kingdom'
}

df['country_clean'] = df['country'].map(canonical)
print(df['country_clean'].tolist())
# ['United States', 'United States', 'United States',
#  'United Kingdom', 'United Kingdom', 'United Kingdom']

空白の除去と大文字・小文字の標準化

import pandas as pd

df = pd.DataFrame({
    'status': ['  Active', 'INACTIVE', 'active ', ' Pending  ', 'ACTIVE']
})

df['status_clean'] = df['status'].str.strip().str.lower()
print(df['status_clean'].value_counts())
# active      3
# inactive    1
# pending     1

タイプミスを修正するファジーマッチング

# Conceptual example (requires: pip install rapidfuzz)
from rapidfuzz import process

canonical_cities = ['New York', 'Los Angeles', 'Chicago', 'Houston']
dirty_values = ['New Yrok', 'Los Angelas', 'Chicagoo']

for dirty in dirty_values:
    best, score, _ = process.extractOne(dirty, canonical_cities)
    print(f'{dirty!r} -> {best!r} (score={score:.0f}%)')
# 'New Yrok'   -> 'New York'    (score=91%)
# 'Los Angelas'-> 'Los Angeles' (score=96%)
# 'Chicagoo'   -> 'Chicago'     (score=94%)

explode()で複数値セルを分割する

import pandas as pd

df = pd.DataFrame({
    'user_id': [1, 2, 3],
    'skills': ['python,sql', 'java,kotlin,sql', 'python']
})

df['skills'] = df['skills'].str.split(',')
exploded = df.explode('skills')
print(exploded)
#    user_id   skills
# 0        1   python
# 0        1      sql
# 1        2     java
# 1        2   kotlin
# 1        2      sql
# 2        3   python

テキストの異なるエンコーディングに対処する

import pandas as pd
import unicodedata

df = pd.DataFrame({'name': ['Caf\u00e9', 'na\u00efve', 'r\u00e9sum\u00e9']})

# Normalize and strip accents (NFKD decomposition + ascii encoding)
df['name_ascii'] = (
    df['name']
    .str.normalize('NFKD')
    .str.encode('ascii', errors='ignore')
    .str.decode('ascii')
)
print(df)
#       name name_ascii
# 0     Cafe       Cafe
# 1    naive      naive
# 2   resume     resume

複合キーを作成する

import pandas as pd

df = pd.DataFrame({
    'first': ['  Alice', 'BOB', ' Carol'],
    'last': ['Smith ', 'JONES', 'White  '],
    'dob': ['1990-01-15', '1985-07-22', '1992-11-30']
})

df['match_key'] = (
    df['first'].str.strip().str.lower() + '|' +
    df['last'].str.strip().str.lower() + '|' +
    df['dob']
)
print(df['match_key'].tolist())
# ['alice|smith|1990-01-15', 'bob|jones|1985-07-22', 'carol|white|1992-11-30']

正規カテゴリリストを適用する

import pandas as pd

df = pd.DataFrame({
    'status': ['active', 'inactive', 'active', 'archived', 'unknown_status']
})

canonical = {'active', 'inactive', 'archived'}

unknown = df[~df['status'].isin(canonical)]['status'].unique()
print('Unknown statuses:', unknown.tolist())
# ['unknown_status']

# Flag unknown rows
df['status_valid'] = df['status'].isin(canonical)
print(df)

完全なテキストクリーニングパイプライン

import pandas as pd

def clean_category_column(series, canonical_map, canonical_set):
    cleaned = (
        series
        .str.strip()
        .str.lower()
        .map(lambda x: canonical_map.get(x, x))  # fix known variants
    )
    unknown = cleaned[~cleaned.isin(canonical_set)]
    if not unknown.empty:
        print('Warning: unknown values:', unknown.unique().tolist())
    return cleaned

cmap = {'eng': 'engineering', 'hr': 'human_resources', 'mktg': 'marketing'}
cset = {'engineering', 'human_resources', 'marketing', 'finance'}

df = pd.DataFrame({'dept': ['Eng', 'HR', 'Marketing', 'MKTG', 'Legal']})
df['dept_clean'] = clean_category_column(df['dept'], cmap, cset)
print(df)

理解度チェック

レッスンのまとめ