Difficulty: Hard | Concepts: Pandas GroupBy, Rolling Window, Transform, Standard Deviation, Time Series

Calculating a rolling average in Pandas is easy on a single series — the real Pandas interview question for data engineers is doing it correctly per group, on real transactional data, and then using that rolling baseline to flag anomalies. This shows up constantly in ETL pipelines, monitoring dashboards, and BI reporting layers.

Pandas Rolling Average — Problem Statement

You are given daily sales data for multiple stores. For each store, calculate the 7-day rolling average of sales using Pandas, then flag any day where the actual sales deviate from that rolling average by more than 2 standard deviations as a potential anomaly.

Dataset

import pandas as pd

data = {
    "store_id": ["A"]*10 + ["B"]*10,
    "date": pd.date_range("2026-06-01", periods=10).tolist() * 2,
    "sales": [200, 210, 195, 205, 215, 198, 202, 500, 208, 199,
              150, 155, 148, 160, 152, 158, 149, 300, 151, 153]
}

df = pd.DataFrame(data)

Expected Output (Sample)

Store	Date	Sales	Rolling Avg (7d)	Is Anomaly
A	2026-06-08	500	203.6	True
A	2026-06-09	208	246.1	False
B	2026-06-08	300	153.1	True

Solution — GroupBy + Rolling Window in Pandas

df = df.sort_values(["store_id", "date"])

# Rolling average and rolling std, calculated separately per store
df["rolling_avg"] = (
    df.groupby("store_id")["sales"]
      .transform(lambda x: x.rolling(window=7, min_periods=3).mean())
)

df["rolling_std"] = (
    df.groupby("store_id")["sales"]
      .transform(lambda x: x.rolling(window=7, min_periods=3).std())
)

# Flag anomalies: sales that fall more than 2 standard deviations
# away from the rolling average
df["is_anomaly"] = (
    (df["sales"] - df["rolling_avg"]).abs() > 2 * df["rolling_std"]
)

print(df[["store_id", "date", "sales", "rolling_avg", "is_anomaly"]])

Explanation

The key idea interviewers are testing here is whether you know to use groupby().transform() instead of apply() or a manual loop. transform() returns a result aligned back to the original DataFrame's index, so the rolling calculation stays correctly scoped per store without mixing data across groups — a very common Pandas groupby rolling window mistake candidates make under interview pressure.

min_periods=3 is added deliberately so the rolling average still produces a value early in each store's date range, instead of returning NaN until a full 7-day window is available — an important edge case in real time series anomaly detection with Pandas.

Visual Representation

Store A daily sales: 200 210 195 205 215 198 202 [500] 208 199 ^ far above rolling average + 2*std --> flagged as anomaly Rolling window slides one day at a time, recalculating mean & std within each store group only.

Conclusion

This pattern — rolling average by group in Pandas combined with a standard deviation based anomaly detection rule — is one of the most practical Python interview questions for data engineers and BI developers, because it mirrors exactly what you'd build for a real sales monitoring or fraud detection pipeline, not just an academic coding puzzle.