25. How to Rank and Running Totals in pandas (rank, cumsum, window)

25.6 Common Mistakes

By Ravindra Bagale, trainer LinkedIn Last updated: 24 October 2026

Why. Silent wrong ranks and cumsums look “professional” in Excel exports until someone audits Thursday.

What. Watch for: unsorted dates; ranking the wrong direction; forgetting groupby; applying rolling on ticket-level rows instead of daily aggregates; chaining filters then ranking a copy you throw away.

How.

Checklist before you send

sort_values on the time key before cumsum / rolling.
Print ascending=True vs False intent (speed vs GMV).
For per-city metrics, confirm groupby key spelling (Pune vs pune).
Aggregate to day grain before multi-day rolling.
After rank, value_counts on the rank column to see ties.
Keep a raw column; add rank/cumsum as new columns.

Worked example A — Broken cumsum (forgot sort).

import pandas as pd

messy = pd.DataFrame({
    "order_date": pd.to_datetime(["2026-10-24", "2026-10-22", "2026-10-23"]),
    "gmv": [140000, 88000, 125000],
})
print("wrong", messy["gmv"].cumsum().tolist())
fixed = messy.sort_values("order_date")
print("right", fixed["gmv"].cumsum().tolist())

Worked example B — Rank direction bug.

mins = pd.Series([18, 11, 25], name="delivery_mins")
# Wrong for "fastest": ranking high minutes first
print("wrong", mins.rank(ascending=False, method="dense").tolist())
print("right", mins.rank(ascending=True, method="dense").tolist())

Worked example C — Rolling on orders grain.

Zoya rolled amount_inr on 5,000 order rows with rolling(3) — that is “last 3 tickets”, not “last 3 days”. Aggregate by date first.

Ravindra Bagale's Tip

Bug checklist: sort, direction, group key, grain. Char shabdat audit.