Pandas — Sorting & Top-N Analysis

• Sort rows by one or many columns; ascending/descending per-column.
• Use nlargest / nsmallest for fast top-N.
• Rank rows with .rank() and the method= option.
• Combine sort + group for "top N per group".

df.sort_values("score", ascending=False)

That's 95% of what you'll ever need. The other 5%: multiple keys, NA placement, sort by index.

Multiple sort keys

df.sort_values(["class", "score"], ascending=[True, False])
# class asc, then within class score desc

nlargest / nsmallest — faster than sort + head

df.nlargest(3, "score")          # top 3 by score
df.nsmallest(3, "score")         # bottom 3
df.nlargest(5, ["score", "age"]) # primary score, tiebreaker age

Internally these don't fully sort the frame — they use partial sorting, which is faster for large data.

Sort by index

df.sort_index()                    # by row index
df.sort_index(axis=1)              # by column name

Rank

df["rank"] = df["score"].rank(ascending=False, method="dense")

method= options: average (default), min, max, first, dense. For leaderboards dense avoids gaps after ties.

Top N per group

# Top 2 students per class
(df.sort_values(["class", "score"], ascending=[True, False])
   .groupby("class").head(2))

Bonus — sort by a derived value

df.iloc[df["name"].str.len().argsort()]   # by length of name

import pandas as pd

df = pd.read_csv("clean.csv", parse_dates=["date"])
df["total"] = df["quantity"] * df["price"]

# 1. Top 5 customers by spend
top_customers = (df.groupby("customer", as_index=False)["total"].sum()
                   .nlargest(5, "total")
                   .reset_index(drop=True))
print(top_customers)

# 2. Worst day by revenue
worst_day = (df.groupby(df["date"].dt.date)["total"].sum()
               .nsmallest(1))
print(f"\n📉 worst day: {worst_day.index[0]}  RM {worst_day.values[0]:.2f}")

# 3. Top 2 products per day
top_per_day = (df.groupby(["date", "product"], as_index=False)["total"].sum()
                 .sort_values(["date", "total"], ascending=[True, False])
                 .groupby("date").head(2))
print("\n🏆 top 2 per day:")
print(top_per_day)

# 4. Add a rank
df["rank_today"] = (df.groupby(df["date"].dt.date)["total"]
                      .rank(ascending=False, method="dense"))

Read the diff

Same DataFrame, four different reports. The combo groupby + sort + head is the Pythonic "top N per group" — easier to read than window functions in SQL once you've used it twice.

sort_values + slicing covers nearly every ordering need. nlargest/nsmallest are faster when you only want N. rank with method="dense" is the leaderboard helper. Combine sort with groupby to compute top-N per group. Tomorrow we level up with the big challenge.