Understanding pandas loc/iloc Performance and Faster Alternatives with at/iat

When working with pandas DataFrames, the loc[] function is used to access values by row label and column name, while iloc[] accesses by integer position. The article begins with a simple example showing how to retrieve the value 10 from column "a" in the second row using df.loc[1, 'a'] and the equivalent df.iloc[1, 0] .

It then illustrates the performance problem that arises when loc (or iloc ) is used inside a Python for loop to create a new column "c" as the sum of columns "a" and "b". The loop iterates over each row with df.iterrows() and assigns the result with df.loc[index, 'c'] = row.a + row.b . The measured execution time is about 40 minutes (2414 seconds), demonstrating that this approach is inefficient.

<code>##df.loc[index, column_name]

df.loc[1,'a']
### 输出：10
</code>

As a faster alternative, the article recommends replacing loc with at (or iloc with iat ) for scalar assignments. Using df.at[index, 'c'] = row.a + row.b reduces the execution time to roughly 40 seconds, a speed‑up of about 60×.

<code>##df.at[index, column_name]

df.at[1,'a']
### 输出：10
</code>

The reason for the speed difference is explained: at / iat are lightweight accessors designed for single scalar values, while loc / iloc support vectorized operations on multiple rows or columns, incurring additional overhead.

In conclusion, the article advises avoiding loc / iloc inside loops for element‑wise updates and instead using at / iat when possible, as they are significantly faster for scalar assignments.