Why Using loc/iloc in Loops Is Slow and How at/iat Provides Faster Alternatives in Pandas

When iterating over a pandas DataFrame, using the loc or iloc accessor to read or write values can take a very long time; the article investigates the cause and presents faster alternatives.

loc[] is a pandas function that accesses DataFrame values by row label and column name. For example, given a DataFrame df , df.loc[1, 'a'] returns the value 10.

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

df.loc[1,'a']

### 输出：10</code>

iloc works similarly but uses integer positions instead of labels, e.g., df.iloc[1,0] also returns 10.

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

df.iloc[1,0]

### 输出：10</code>

Because loc accesses by column name and iloc by column index, they are suitable for vectorized operations on whole rows or columns, not for scalar updates inside loops.

Example: adding a new column c as the sum of columns a and b using a for loop with loc :

<code>import timestart = time.time()

# 遍历 DataFrame df 
for index, row in df.iterrows():
        df.loc[index, 'c'] = row.a + row.b

end = time.time()
print (end - start)

### 所用时间：2414 秒</code>

This loop takes about 40 minutes.

Replacing loc with the scalar accessor at (or iloc with iat ) yields a dramatic speedup:

<code>import timestart = time.time()

# 遍历 DataFrame 
for index, row in df.iterrows():
    df.at[index, 'c'] = row.a + row.b

end = time.time()
print (end - start)

### 所用时间：40 秒</code>

The at / iat version runs in about 0.7 minutes, roughly 60 times faster than the loc version.

The reason is that at and iat are designed for scalar access, making them lightweight and fast, whereas loc and iloc are built for accessing Series or DataFrames and involve additional overhead.

Therefore, using loc / iloc inside Python loops is not recommended; for element‑wise updates, prefer at / iat to achieve much better performance.