Pandas resample: choose time-bin boundaries and preserve empty hours

Operation contract

The receipt series uses sorted UTC timestamps and one-hour bins closed and labeled on the left. Two receipts in the first hour sum to 375. The empty next hour remains missing because min_count is one. A zero amount in a populated hour would be a real zero, which is different from no observed receipt.

Failure and ownership boundary

UTC avoids ambiguous local-clock bins in this fixture. Business-day reporting may instead require a named timezone, conversion and an explicit daylight-saving policy. Resampling does not automatically remove duplicate records or establish arrival completeness; Python zoneinfo and fold: two instants can share one clock label and Pandas rolling windows: minimum observations and causal boundaries cover separate timing problems.

Tested environment

Dependency check: this program was executed on CPython 3.14.6 with pandas==3.0.6. Install these versions in a separate virtual environment. The download includes the recorded environment snapshot; no third-party package is part of the website runtime.

Working program

import pandas as pd

index = pd.to_datetime(["2026-09-01T00:10:00Z", "2026-09-01T00:50:00Z", "2026-09-01T02:10:00Z"])
amounts = pd.Series([125, 250, 75], index=index, dtype="Int64")
if not amounts.index.is_monotonic_increasing or amounts.index.hasnans:
    raise ValueError("ordered timestamps required")
hourly = amounts.resample("1h", closed="left", label="left").sum(min_count=1)
print([timestamp.strftime("%H:%M") for timestamp in hourly.index])
print([None if pd.isna(amount) else int(amount) for amount in hourly])

Output

['00:00', '01:00', '02:00']
[375, None, 75]

Costs and limits

Result storage depends on the number of generated bins, not only the number of observations. Sparse events separated by a very long time interval can create many empty bins. Bound both input count and reporting horizon in a receiving application.

Common Mistakes

• An empty time bin is not a measured zero.
• Declare endpoint and timezone policies instead of relying on implicit reporting assumptions.

Connected lessons

Pandas rolling windows: minimum observations and causal boundaries, Python zoneinfo and fold: two instants can share one clock label, Pandas groupby: retain missing keys and define all-null totals.

Follow the service contract

Python time-series validation: fit on past rows and leave a declared gap.