fillna / dropna

fillna

result = leads.fillna({"city": "unknown", "score": 0})

SELECT lead_id, name,
       coalesce(city, 'unknown') AS city,
       coalesce(score, 0)        AS score,
       phone
FROM leads

Each column gets its own default, and columns not in the dict keep their nulls. SQL has no fillna: coalesce(col, default) per column is the equivalent.

The type rule

fillna(0) fills only numeric columns, and fillna("n/a") fills only string columns. Nothing errors when a column is skipped, so leads.fillna(0) leaves city null and it is easy to think everything was filled. A dict makes the intent explicit and avoids the surprise.

dropna

result = leads.dropna(thresh=3)

SELECT * FROM leads
WHERE (CASE WHEN lead_id IS NOT NULL THEN 1 ELSE 0 END
     + CASE WHEN name    IS NOT NULL THEN 1 ELSE 0 END
     + CASE WHEN city    IS NOT NULL THEN 1 ELSE 0 END
     + CASE WHEN score   IS NOT NULL THEN 1 ELSE 0 END
     + CASE WHEN phone   IS NOT NULL THEN 1 ELSE 0 END) >= 3

thresh overrides how. subset limits which columns are checked, and it combines with both.

When not to fill

A null often carries meaning: "not measured", "not applicable", "not yet known". Filling a missing score with 0 makes avg(score) lower, because avg ignores nulls but not zeros. Fill for display and for columns where a default is truly correct (a missing quantity in a cart is 0); keep nulls in data used for statistics.

Related tools

• coalesce(a, b, c) fills from other columns, not a constant: see the coalesce lesson.
• df.na.replace(["N/A", ""], None, subset=["city"]) turns placeholder strings into real nulls first, so fillna and dropna see them.
• For a forward fill (use the last known value), use last(col, ignorenulls=True) over a window.

Common mistakes