Level 9 · Pick a welcome offer per customer

apply on a Series: one value at a time

Series.apply(fn) calls your function once per value and collects the answers into a new Series. A def is the right shape when the rule has several branches, and it’s where you deal with missing values:

def source_group(src):
    if pd.isna(src):
        return "Unknown"
    if src in ("instagram", "tiktok", "pinterest"):
        return "Social"
    if src in ("search", "google_ads"):
        return "Search"
    return "Other"

customers["referral_source"].apply(source_group).value_counts()
# Unknown 666, Search 548, Social 474, Other 330

A missing referral source shows up in pandas as NaN. NaN isn’t equal to anything, not even itself, so src == None never catches it. pd.isna(src) does, and it works on single values as well as on whole columns. Put that check first, before any test that assumes text.

For a one-line rule, a lambda saves the def:

products["unit_price"].apply(lambda p: round(p * 1.03, 2))   # a 3% price rise

That one is really vectorised maths in disguise: (products["unit_price"] * 1.03).round(2) does the same and runs far faster. Reach for apply when the logic needs if.

apply on rows: axis=1

Jordan’s offer depends on two columns, segment and the new source_group. With axis=1, DataFrame.apply passes your function one whole row at a time, and you read any column from it by name:

def offer(row):
    if row["segment"] == "business":
        return "account call"
    if row["source_group"] == "Social":
        return "creator code"
    if row["source_group"] == "Unknown":
        return "survey"
    return "newsletter"

customers["offer"] = customers.apply(offer, axis=1)

The order of the ifs is the business rule: business accounts get a call whatever their source, so that test comes first. 286 business customers get a call, 746 consumers the newsletter, 575 the survey and 411 a creator code.

Which group is biggest? idxmax

max() tells you the largest value; idxmax() tells you where it is, meaning its index label:

counts = customers["source_group"].value_counts()
counts.max()      # 666
counts.idxmax()   # 'Unknown'
counts.drop("Unknown").idxmax()   # 'Search'

That’s a useful reality check: the biggest “source” is not knowing the source at all. Among known sources, search brings in the most customers. On a DataFrame, idxmax() returns the label of the top row for each column.

Common mistakes

• Checking src == None instead of pd.isna(src), so missing values fall through to ‘Other’.
• Forgetting axis=1, so the function receives columns instead of rows and fails.
• Calling the function yourself, apply(source_group()), instead of passing it by name.
• Putting the business test last, so business customers from social get a creator code.
• Using max() when you need the label of the biggest group: that’s idxmax().