Time series with pandas

Once your DataFrame has a DatetimeIndex, a whole set of superpowers unlocks.

Building a DatetimeIndex

pd.date_range("2026-01-01", periods=365, freq="D")
pd.to_datetime(df["ts"])
df.set_index("ts", inplace=True)

Resampling — change the frequency

• .resample("W") — weekly. "W-SUN" = weeks ending Sunday.
• .resample("MS") — month start. "ME" — month end.
• .resample("Q") / "Y" — quarterly / yearly.
• Chain an aggregation: .sum(), .mean(), .agg([...]), or .ohlc() (open/high/low/close — great for prices).

Downsample: coarser (daily → weekly). Upsample: finer (daily → hourly) — usually with .ffill() / .interpolate().

Rolling windows

s.rolling(7).mean()        # 7-day moving average
s.rolling("7D").mean()     # 7 calendar days (respects gaps)
s.rolling(30, min_periods=1).max()   # 30-period max, no leading NaN

Shift, diff, pct_change

• .shift(1) — lag by one period.
• .diff() — s - s.shift(1), absolute change.
• .pct_change() — percentage change (great for growth rates).

Timezones

• .tz_localize("Asia/Kolkata") — attach a tz to naive timestamps.
• .tz_convert("UTC") — convert between tzs.
• Always store UTC in databases; convert on display.

Try it

1. Compute a year-over-year growth rate on monthly data.
2. Find the 7-day rolling max and label peaks where max == today.