The profile an ad network would sell

Sam never told AdCo anything. Sam just shopped. But AdCo saw every page, and each page has a category. Count the categories and a picture appears: someone into running, perhaps expecting a baby, perhaps worried about money.

Some categories are more sensitive than others. Health, pregnancy, quitting smoking, money trouble and baby things can reveal a medical condition, a pregnancy or a family's finances, often before the person has told anyone close to them.

The practice world has its own list, SENSITIVE_CATEGORIES, which leaves out baby. This lab's SENSITIVE set adds it, because baby shopping so often hints at a pregnancy.

Counting with a dictionary

counts = {}
for category in ["running", "baby", "running"]:
    counts[category] = counts.get(category, 0) + 1
print(counts)   # {'running': 2, 'baby': 1}

counts.get(key, 0) means "the current count, or 0 if this is new". Python also has collections.Counter, which does the same job in one line; either is fine.

To find the key with the biggest count, max(counts, key=counts.get) looks at every key but compares them by counts.get(key), their count:

print(max({"running": 2, "baby": 1}, key={"running": 2, "baby": 1}.get))   # running

In this lab you use a longer session (40 pages) with a normal browser. First find AdCo's main ID for Sam (the cookie ID seen on the most records), then count the categories of that ID's records, then list which sensitive categories appear.

The defence

Partitioning (lesson 41) cuts this profile into one small piece per shop, and a blocker (Module 8) stops the reports entirely. Neither changes what each shop knows about its own visitors; they change what one outsider can join together.

const counts = {};
for (const r of mine) counts[r.category] = (counts[r.category] ?? 0) + 1;

var counts = mine.GroupBy(r => r.Category).ToDictionary(g => g.Key, g => g.Count());