Level 9 · A repeatable spot-check sample

A filter result is not always a separate table

When you filter, pandas may hand you something that is still tied to the original DataFrame. If you then add or change a column on it, pandas can’t be sure whether you meant to change orders too, and it warns you:

web = orders[orders["channel"] == "web"]
web["free_shipping"] = web["shipping_fee"] == 0
# SettingWithCopyWarning: A value is trying to be set on a copy of a slice from a DataFrame.

The warning is pandas saying “this might not do what you think”. Here the column lands only in web, but closely related code, such as orders[orders["channel"] == "web"]["shipping_fee"] = 0, silently changes nothing at all. The clean answer is to say what you want: a separate table of your own.

web = orders[orders["channel"] == "web"].copy()
web["free_shipping"] = web["shipping_fee"] == 0   # no warning, orders is untouched

Plain assignment, web2 = web, doesn’t copy anything either. It gives the same table a second name, so a change through one name shows up through the other. Use .copy() whenever you plan to modify a filtered or borrowed table.

Random rows you can reproduce

sample picks rows at random. n says how many; frac=0.1 would take 10% instead.

web.sample(n=10)                   # a different ten every run
web.sample(n=10, random_state=7)   # the same ten every run

random_state is a seed: any whole number works, and the same number on the same table always picks the same rows. That matters for Angela’s team. If someone asks “why did you call these customers?”, you can rerun the code and get the identical list. With random_state=7, the first picks are orders 2453, 12392 and 8576.

sample returns the rows in the random order it drew them, with their original index labels. For a call list that’s fine, so leave the order alone here.

Reading the sample

Eight of the ten sampled orders shipped free. Across all 5,913 web orders the share is about 79%, so the sample looks typical. A small sample can easily look unusual by chance, which is worth knowing before anyone reads too much into ten rows.

Common mistakes

• Skipping .copy() and getting a SettingWithCopyWarning when you add the column.
• Thinking web2 = web makes a copy. It doesn’t; both names point to the same table.
• Leaving out random_state, so the list changes every time you run it.
• Sampling orders first and filtering afterwards, which gives fewer than ten web orders.
• Sorting or resetting the sample when the task asks for it as drawn.