删除缺失值
使用 dropna() 删除包含 NaN 的行或列，并控制阈值以及要检查的列子集。
删除缺失值 是 CoddyKit 上的免费 Pandas & NumPy Academy 课时。 这是第 2 节课，共 4 节。 你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。 这是 Pandas & NumPy Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 Pandas & NumPy Academy 课程共包含 4 节课。
本课时的部分内容尚未翻译，以英文显示。
When to Drop Missing Values?
Dropping rows with missing values is the simplest imputation strategy, but it is only valid when the data is Missing Completely At Random (MCAR) — meaning the probability of a value being missing has nothing to do with the missing value itself or any other variable. If missing data is systematic (e.g., low-income respondents skip the salary field), dropping it introduces bias. Always investigate the missingness pattern before deciding to drop.
import pandas as pd
import numpy as np
# Example: randomly missing salary data (MCAR-like)
df = pd.DataFrame({
'name': ['Alice', 'Bob', 'Carol', 'Dave'],
'salary': [50000, np.nan, 70000, np.nan]
})
print('Before drop:', df.shape) # (4, 2)
print(df)dropna() — Basic Usage
DataFrame.dropna() removes any row that contains at least one NaN value by default. It returns a new DataFrame; the original is unchanged unless you pass inplace=True. For small to medium datasets this default behaviour is often acceptable as a quick data cleaning first pass.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'a': [1, np.nan, 3, np.nan],
'b': [10, 20, np.nan, 40],
'c': [100, 200, 300, 400]
})
cleaned = df.dropna()
print(cleaned)
# a b c
# 0 1.0 10.0 100
print('Original shape:', df.shape) # (4, 3)
print('Cleaned shape:', cleaned.shape) # (1, 3)how='all' — Drop Only All-NaN Rows
Passing how='all' tells dropna to remove a row only if every single value in that row is NaN. This is much less aggressive than the default how='any'. Use how='all' when your dataset has sparse rows — rows that have some data are worth keeping, while entirely empty rows are clearly junk records.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'a': [1, np.nan, np.nan],
'b': [2, np.nan, np.nan],
'c': [3, 4, np.nan]
})
# Row 2 has ALL NaN — dropped
# Row 1 has partial NaN — kept with how='all'
cleaned = df.dropna(how='all')
print(cleaned)
# a b c
# 0 1.0 2.0 3.0
# 1 NaN NaN 4.0subset= — Check Only Specific Columns
The subset parameter limits which columns are checked for NaN when deciding whether to drop a row. This is extremely useful when only certain columns are critical — for example, a row should be dropped if the user_id or target column is missing, but NaN in optional feature columns is acceptable.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'user_id': [1, np.nan, 3],
'score': [88, 95, np.nan],
'notes': ['ok', 'good', np.nan]
})
# Drop row only if user_id is missing — score and notes NaN are ok
cleaned = df.dropna(subset=['user_id'])
print(cleaned)
# user_id score notes
# 0 1.0 88.0 ok
# 2 3.0 NaN NaNthresh= — Minimum Non-Null Requirement
The thresh parameter keeps a row only if it has at least thresh non-NaN values. This is more nuanced than how='any' or how='all': you can say 'keep a row if at least 3 out of 5 columns have data'. This is useful for datasets where some sparsity is expected but completely empty rows should be removed.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'a': [1, np.nan, np.nan],
'b': [2, 3, np.nan],
'c': [4, np.nan, np.nan],
'd': [5, 6, np.nan]
})
# Keep rows with at least 3 non-null values
cleaned = df.dropna(thresh=3)
print(cleaned)
# a b c d
# 0 1.0 2.0 4.0 5.0
# 1 NaN 3.0 NaN 6.0Dropping Columns Instead of Rows
By default, dropna() removes rows (axis=0). Pass axis=1 (or axis='columns') to drop columns that contain any NaN instead. This is appropriate when a column is mostly empty and provides little signal — keeping it would just add noise to a model or summary table.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'id': [1, 2, 3],
'name': ['A', 'B', 'C'],
'temp': [np.nan, np.nan, np.nan], # completely empty column
'score': [80, 90, 85]
})
# Drop columns that have any NaN
cleaned = df.dropna(axis=1)
print(cleaned)
# id name score
# 0 1 A 80
# 1 2 B 90
# 2 3 C 85Dropping Columns by Missing Threshold
A powerful pattern is to drop columns that exceed a certain missing percentage. Compute the fraction missing per column, identify columns above your threshold (e.g., 50%), and drop them with df.drop(columns=cols_to_drop). This is more targeted than dropna(axis=1) which drops any column with even one NaN.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'a': [1, 2, np.nan, 4, 5],
'b': [np.nan, np.nan, np.nan, np.nan, 5], # 80% missing
'c': [1, np.nan, 3, 4, 5] # 20% missing
})
threshold = 0.5
high_missing = df.columns[df.isna().mean() > threshold]
print('Dropping:', high_missing.tolist()) # ['b']
cleaned = df.drop(columns=high_missing)
print(cleaned)Preserving the Index After dropna()
After calling dropna(), the original row indices are preserved — so if rows 1 and 3 were dropped, the remaining DataFrame has indices 0, 2, 4. This is often desirable (you can trace back to original positions), but sometimes you want a clean sequential index starting from 0. Call .reset_index(drop=True) after dropping to renumber rows.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'val': [1, np.nan, 3, np.nan, 5]
})
cleaned = df.dropna()
print('With original index:')
print(cleaned) # indices 0, 2, 4
cleaned_reset = cleaned.reset_index(drop=True)
print('With reset index:')
print(cleaned_reset) # indices 0, 1, 2dropna() in a Pipeline
Since dropna() returns a DataFrame, it integrates naturally into a method chain. Chaining dropna() between load and analysis steps is a clean pattern that keeps the pipeline readable without temporary variables. You can also chain it with query(), assign(), and groupby().
import pandas as pd
import numpy as np
df = pd.DataFrame({
'region': ['East', 'West', None, 'East'],
'revenue': [100, np.nan, 300, 400]
})
result = (
df
.dropna(subset=['region', 'revenue'])
.groupby('region')['revenue'].sum()
)
print(result)
# region
# East 500.0
# dtype: float64When NOT to Drop: Prefer Filling
Dropping rows loses data. For columns with fewer than 5-10% missing values, filling (imputing) is usually better than dropping. Also, if missing values are correlated with the target variable (Missing Not At Random, MNAR), dropping them introduces bias. As a rule: only drop when missingness is truly random, the dataset is large enough that lost rows don't matter, and the column or row provides no recoverable signal.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'user': ['Alice', 'Bob', 'Carol', 'Dave'],
'age': [25, np.nan, 30, np.nan]
})
missing_pct = df['age'].isna().mean()
print(f'Missing age: {missing_pct:.0%}') # 50%
# 50% missing is high — consider imputing instead of dropping
# df['age'].fillna(df['age'].median(), inplace=True)Practical Cleaning Workflow
A practical missing-value workflow combines multiple dropna strategies: first drop entirely empty rows, then drop columns that are more than 60% empty, then drop rows missing critical ID or target columns, and finally fill the remaining scattered NaN values. This layered approach preserves as much data as possible.
import pandas as pd
import numpy as np
df = pd.DataFrame({
'id': [1, 2, 3, np.nan],
'feature': [1.0, np.nan, 3.0, 4.0],
'useless': [np.nan, np.nan, np.nan, np.nan]
})
clean = (
df
.dropna(how='all') # remove all-NaN rows
.drop(columns=df.columns[df.isna().mean() > 0.9]) # remove >90% empty cols
.dropna(subset=['id']) # must have an ID
)
print(clean)
# id feature
# 0 1.0 1.0
# 1 2.0 NaN
# 2 3.0 3.0Quick Check
Test your understanding of dropping missing values with dropna().
Lesson Recap
In this lesson you learned: dropna() removes rows with NaN by default, how='all' only drops fully-empty rows, subset= restricts checking to specific columns, and thresh= keeps rows with a minimum number of non-null values. Use axis=1 to drop columns instead of rows. Next up we fill missing values instead of dropping them using fillna().
常见问题解答
「删除缺失值」课时是免费的吗？
是的 — 「删除缺失值」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 Pandas & NumPy Academy 课程的其余内容，请升级到 CoddyKit PRO。 Pandas & NumPy Academy 课程共包含 4 节课。
「删除缺失值」这节课中我会学到什么？
使用 dropna() 删除包含 NaN 的行或列，并控制阈值以及要检查的列子集。 你通过在浏览器中直接运行的动手代码来练习 Pandas & NumPy Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。
学习 Pandas & NumPy Academy 需要有经验吗？
无需任何先前经验。CoddyKit 上的 Pandas & NumPy Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。 这是第 2 节课，共 4 节。
「删除缺失值」课时需要多长时间？
大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。
我能在这节 Pandas & NumPy Academy 课中编写并运行代码吗？
能。每节 Pandas & NumPy Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。
此课程中的所有课时
- 检测缺失值
- 删除缺失值
- 填充缺失值
- 插值与高级缺失值填补