Python · medium

Given a pandas DataFrame, compute the 30-day moving average of a column.

Asked in Data Scientist interviews, in the Python round.

Short answer

df['ma'] = df['x'].rolling(30).mean().

Sample data

These tables are loaded into an in-browser engine on this page, so you can run your own answer against them. No account needed.

prices

How to answer it

Definition

The average of a column over the current row and the 29 before it. In pandas that is .rolling(), and the question is really about what "30" counts: rows, or days.

Syntax

df["ma_30"] = df["value"].rolling(window=30).mean()          # 30 ROWS
df["ma_30"] = df["value"].rolling("30D").mean()              # 30 DAYS

Example

import pandas as pd

df = df.sort_values("date").set_index("date")

# One row per day first, so a row and a day mean the same thing
daily = df["revenue"].resample("D").sum()
ma = daily.rolling("30D").mean()
print(ma.tail(3))

Output

date
2025-03-29    41832.4
2025-03-30    41905.1
2025-03-31    42017.8
Freq: D, Name: revenue, dtype: float64

Rows or days

rolling(30) counts 30 rows. If the frame has gaps, or several rows per day, that is not 30 days. rolling("30D") counts a time offset and needs a DatetimeIndex, which is why the example sets one. Resampling to a daily grain first makes the two equivalent and removes the ambiguity entirely.

Key points

• Sort by the date first. rolling trusts the current order and will happily average nonsense.
• rolling(30) is rows, rolling("30D") is days. Say which the question wants.
• The first 29 values are NaN by design. min_periods=1 fills them with partial averages, which is a decision rather than a default.
• .rolling() includes the current row. For a trailing average that excludes today, .shift(1) first.
• Grouped version: df.groupby("user")["value"].rolling(30).mean(), then reset the index.

FAQ

• Why are my results all NaN? Fewer than 30 rows, or NaNs inside the window. Check min_periods.
• Rolling or expanding? Rolling has a fixed window; .expanding() grows from the start.
• Is this leakage in a model? Yes, if the window includes the current row and the target derives from it. Shift before using it as a feature.

The 30-second answer

Sort by date, then rolling thirty and take the mean.

The thing I would clarify before writing it is whether thirty means thirty rows or thirty days, because rolling with a number counts rows. If the data has gaps, or more than one row per day, those are different answers, so I would either resample to a daily grain first or use the offset form with "30D", which needs a DatetimeIndex.

I would also mention that the first twenty-nine values come back NaN by design, and that filling them through min_periods is a choice to make deliberately rather than to default into.

In the interview

• What they are checking: rows versus days, and the sort.
• Common mistake: not sorting first. A rolling window over unsorted dates produces numbers that look plausible and mean nothing.

Practice this for real

• Check your resume against a Data Scientist job description, free, no account needed.
• Every data interview question, by role and topic
• Scored AI mock interviews, two free, no card