결측값 채우기
fillna()와 method 매개변수를 사용해 NaN을 상수, 열 평균, 앞의 값 또는 뒤의 값으로 대체합니다.
결측값 채우기은(는) CoddyKit의 무료 Pandas & NumPy Academy 강의입니다. 이것은 4개 중 3번째 강의입니다. 아래에서 전체 강의를 무료로 읽을 수 있으며, 내장 코드 에디터와 24/7 AI 튜터와 함께 브라우저에서 직접 실습할 수 있습니다. 이 강의는 Pandas & NumPy Academy 학습 경로의 일부이며, 진행 상황이 웹과 CoddyKit 앱에 동기화됩니다. Pandas & NumPy Academy 강의에는 총 4개의 강의가 포함되어 있습니다.
이 강의의 일부는 아직 번역되지 않았으며 영어로 표시됩니다.
Introduction to fillna()
Imputation means replacing missing values with a reasonable substitute rather than dropping the row. Pandas fillna() is the primary tool for this: it replaces every NaN in a Series or DataFrame with a value you specify. Unlike dropping, imputation preserves all rows, which is especially important when data is scarce or when the missing-data pattern is informative.
The simplest form passes a scalar: df['col'].fillna(0).
import pandas as pd
import numpy as np
df = pd.DataFrame({
'product': ['A', 'B', 'C', 'D'],
'price': [10.0, np.nan, 30.0, np.nan]
})
# Fill all NaN in 'price' with 0
filled = df['price'].fillna(0)
print(filled)
# 0 10.0
# 1 0.0
# 2 30.0
# 3 0.0Filling with a Column Mean or Median
Filling with the column mean (for symmetric distributions) or median (for skewed distributions) is one of the most common imputation strategies. These statistics represent the central tendency of the data, so they minimise the distortion of the column's distribution. Always compute the statistic on the training split to avoid data leakage.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'salary': [50000, np.nan, 70000, 80000, np.nan, 60000]
})
mean_salary = df['salary'].mean()
median_salary = df['salary'].median()
df['salary_mean_filled'] = df['salary'].fillna(mean_salary)
df['salary_median_filled'] = df['salary'].fillna(median_salary)
print(df)
# salary salary_mean_filled salary_median_filled
# 0 50000.0 50000.0 50000.0
# 1 NaN 65000.0 65000.0
# 2 70000.0 70000.0 70000.0Filling Categoricals with Mode
For categorical columns, filling with the mean or median doesn't make sense. Instead, use the mode — the most frequently occurring value. Series.mode()[0] returns the most common value (the [0] handles the case where there are multiple modes).
import pandas as pd
import numpy as np
df = pd.DataFrame({
'status': ['active', 'inactive', None, 'active', None, 'active']
})
most_common = df['status'].mode()[0]
print('Mode:', most_common) # active
df['status'] = df['status'].fillna(most_common)
print(df['status'].tolist())
# ['active', 'inactive', 'active', 'active', 'active', 'active']Forward Fill with ffill()
Forward fill (also called last-observation-carried-forward, LOCF) propagates the last valid (non-NaN) value forward to fill subsequent NaN positions. It is ideal for time series data where a measurement stays constant until updated — for example, device status, price levels, or sensor readings that only change at specific events.
import pandas as pd
import numpy as np
price = pd.Series(
[100, np.nan, np.nan, 105, np.nan, 110],
index=pd.date_range('2024-01', periods=6, freq='ME')
)
filled = price.ffill()
print(filled)
# 2024-01-31 100.0
# 2024-02-29 100.0
# 2024-03-31 100.0
# 2024-04-30 105.0
# 2024-05-31 105.0
# 2024-06-30 110.0Backward Fill with bfill()
Backward fill (next-observation-carried-backward, NOCB) propagates the next valid value backward to fill preceding NaN positions. This is useful when you know future data fills in missing past values — for example, if you receive end-of-month data and need to back-fill days in that month before the report arrives.
import pandas as pd
import numpy as np
df = pd.DataFrame({'value': [np.nan, np.nan, 3, np.nan, 5]})
print(df['value'].bfill())
# 0 3.0
# 1 3.0
# 2 3.0
# 3 5.0
# 4 5.0
# dtype: float64limit Parameter for Forward and Backward Fill
Both ffill() and bfill() accept a limit parameter that restricts how many consecutive NaN values are filled. This is important when you want to carry a value forward only for a short gap — long gaps should remain NaN to signal that the data is genuinely missing, not just delayed.
import pandas as pd
import numpy as np
s = pd.Series([1.0, np.nan, np.nan, np.nan, 5.0])
# Fill only the first NaN gap, leave the rest
print(s.ffill(limit=1))
# 0 1.0
# 1 1.0 <- filled
# 2 NaN <- still NaN (limit reached)
# 3 NaN
# 4 5.0Filling Different Columns Differently
In a real DataFrame, different columns may need different fill strategies. Pass a dictionary to fillna() where keys are column names and values are the fill values. This applies column-specific imputation in a single call, which is cleaner than chaining multiple individual fillna calls.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'age': [25, np.nan, 30],
'income': [50000, np.nan, 70000],
'country': ['USA', np.nan, 'UK']
})
fill_values = {
'age': df['age'].median(),
'income': df['income'].mean(),
'country': 'Unknown'
}
filled = df.fillna(fill_values)
print(filled)
# age income country
# 0 25.0 50000.0 USA
# 1 27.5 60000.0 Unknown
# 2 30.0 70000.0 UKGroup-Aware Imputation
A more sophisticated strategy is to fill NaN values with the group mean or median rather than the overall column mean. For example, fill a missing salary with the median salary for that job category. This preserves subgroup structure and produces more realistic imputations than a global statistic.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'dept': ['Eng', 'Eng', 'HR', 'HR', 'Eng'],
'salary': [90000, np.nan, 50000, np.nan, 110000]
})
# Fill salary NaN with the median salary for each department
df['salary'] = df.groupby('dept')['salary'].transform(
lambda x: x.fillna(x.median())
)
print(df)
# dept salary
# 0 Eng 90000.0
# 1 Eng 100000.0 <- group median (90k+110k)/2
# 2 HR 50000.0
# 3 HR 50000.0
# 4 Eng 110000.0Filling with a Constant Sentinel Value
Sometimes the right imputation is a sentinel value that signals 'unknown' rather than a real measurement. For example, -1 for unknown age, 'N/A' for unknown category, or False for an unknown boolean flag. Sentinel values are valid when downstream code explicitly checks for them and treats them differently from real data.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'user_id': [1, 2, 3],
'referral_source': ['email', np.nan, 'ad'],
'trial_days': [14, np.nan, 7]
})
df['referral_source'] = df['referral_source'].fillna('unknown')
df['trial_days'] = df['trial_days'].fillna(-1).astype(int)
print(df)
# user_id referral_source trial_days
# 0 1 email 14
# 1 2 unknown -1
# 2 3 ad 7Chaining fillna() in a Pipeline
Like all Pandas methods, fillna() returns a new DataFrame and integrates cleanly into a method chain. Calling .fillna() after .dropna() applies a two-step strategy: drop rows missing critical columns, then fill remaining optional NaN values with sensible defaults — all in one readable pipeline.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'user_id': [1, None, 3, 4],
'score': [88, 95, np.nan, 76],
'label': ['A', 'B', None, 'D']
})
cleaned = (
df
.dropna(subset=['user_id'])
.fillna({'score': df['score'].mean(), 'label': 'Unknown'})
)
print(cleaned)
# user_id score label
# 0 1.0 88.0 A
# 2 3.0 86.3 Unknown
# 3 4.0 76.0 DVerifying No NaN Remains
After imputation, always verify that no NaN values remain in the columns you targeted. Call df.isna().sum() or assert that the count is zero. An unexpected non-zero count means your fillna was missing a case — perhaps a column had a dtype that prevented the fill, or you forgot a column in your dict.
import pandas as pd
import numpy as np
df = pd.DataFrame({'a': [1.0, np.nan, 3.0], 'b': [np.nan, 2.0, 3.0]})
filled = df.fillna(0)
# Verify
assert filled.isna().sum().sum() == 0, 'Some NaN remain!'
print('All NaN filled successfully')
print(filled.isna().sum())
# a 0
# b 0
# dtype: int64Quick Check
Test your understanding of filling missing values in Pandas.
Lesson Recap
In this lesson you learned: fillna(scalar) replaces all NaN with a constant, fillna(mean/median) imputes with column statistics, and ffill()/bfill() propagate adjacent values for time series. Pass a dictionary to fillna() for column-specific strategies, and use groupby().transform() for group-aware imputation. Next up we cover interpolation and when to choose advanced imputation techniques.
자주 묻는 질문
“결측값 채우기” 강의는 무료인가요?
네 — “결측값 채우기” 전체 내용을 이 웹사이트에서 무료로 읽을 수 있습니다. 인터랙티브하게 실습하려면(내장 코드 에디터와 24/7 AI 튜터), CoddyKit PRO로 업그레이드하면 Pandas & NumPy Academy 강의 전체를 잠금 해제할 수 있습니다. Pandas & NumPy Academy 강의에는 총 4개의 강의가 포함되어 있습니다.
“결측값 채우기”에서 뭘 배우나요?
fillna()와 method 매개변수를 사용해 NaN을 상수, 열 평균, 앞의 값 또는 뒤의 값으로 대체합니다. 브라우저에서 직접 실행하는 실습 코드로 Pandas & NumPy Academy을(를) 배우며, 24/7 AI 튜터가 강의를 진행하면서 질문에 답변해줍니다.
Pandas & NumPy Academy을(를) 시작하는 데 경험이 필요한가요?
사전 경험은 필요하지 않습니다. CoddyKit의 Pandas & NumPy Academy은(는) 초급자부터 고급 학습자까지를 위해 구성되어 있으므로, 여기서 시작하거나 처음부터 시작할 수 있으며 자신의 속도대로 진행할 수 있습니다. 이것은 4개 중 3번째 강의입니다.
“결측값 채우기” 강의는 얼마나 걸리나요?
대부분의 CoddyKit 강의는 약 5~10분이 소요됩니다. 각 강의는 간결하고 인터랙티브하여 꾸준한 진행이 가능하며, 웹과 앱에서 중단한 부분부터 바로 시작할 수 있습니다.
이 Pandas & NumPy Academy 강의에서 코드를 작성하고 실행할 수 있나요?
네. 모든 Pandas & NumPy Academy 강의에는 내장 코드 에디터가 포함되어 있으므로, 브라우저에서 바로 실제 코드를 작성하고 실행한 후 즉시 AI 피드백을 받을 수 있습니다 — 로컬 설정이 필요 없습니다.