Level 1 · Orders per month with resample

Grouping by time

groupby groups rows that share a value. Time needs something slightly different: you want every order from 3 March 2025 at 14:07 and every order from 28 March at 21:50 to land in the same bucket, “March 2025”. resample does exactly that. It is a groupby for time: it cuts the timeline into regular periods (days, weeks, months, quarters, years) and groups the rows by the period they fall into.

It needs the timestamps in the index, so the pattern always starts with set_index:

monthly = orders.set_index("order_date").resample("MS").size()
monthly.head(3)
# order_date
# 2023-01-01    192
# 2023-02-01    157
# 2023-03-01    191

After resample you call an aggregation, exactly as after groupby: .size() counts rows, ["shipping_fee"].sum() adds a column, .mean(), .max() and friends all work.

The frequency codes

The string you pass is a frequency alias. The ones analysts use most:

alias	period	label
"D"	day	the day
"W-SUN" (or "W")	week ending Sunday	the Sunday
"MS"	calendar month	first day of the month
"ME"	calendar month	last day of the month
"QS" / "QE"	quarter	first / last day
"YS" / "YE"	year	first / last day

"MS" and "ME" produce the same counts; only the label differs. January 2023 is 2023-01-01 with "MS" and 2023-01-31 with "ME". Pick the one that matches whatever you’ll join or compare against. Here Megan’s targets use the first of the month, so "MS" it is. (Older tutorials write "M"; recent pandas versions have renamed it "ME" and warn about the old name.)

What the result looks like

Harbor & Pine’s orders span 36 months. The quietest was February 2023 with 157 orders; the busiest was December 2025 with 777. Every November and December is roughly double a normal month, which you’ll see straight away when you scan the column.

The result is a Series indexed by month. To share it as a table, name the values and move the index back into a column:

monthly.rename("n_orders").reset_index()
#     order_date  n_orders
# 0   2023-01-01       192
# 1   2023-02-01       157

Why not group by month text?

You could build a "2025-03" text column with dt.strftime and groupby it. That works, but resample has two advantages. The result keeps real dates, so you can plot it, shift it and compute growth on it. And a month with no orders still appears, with a count of 0, instead of silently vanishing from the table, which matters as soon as you compute “change since last month”.

Common mistakes

• Calling orders.resample("MS") without set_index. pandas raises a TypeError asking for a DatetimeIndex (or pass on="order_date" instead).
• Using "ME" when the other table is keyed by month starts, or the reverse. The numbers are right but nothing lines up when you merge.
• Using .count() on the whole frame, which returns one count per column. .size() returns one number per month.
• Writing the deprecated "M" alias from older code.