10 Pandas Tricks That Will Save You Hours

Pandas is so flexible that most people use 10% of it badly. These ten patterns will clean up almost any notebook.

1. Method chaining beats temporary variables

# Instead of:
df2 = df[df.country == "BR"]
df2["revenue_k"] = df2.revenue / 1000
result = df2.groupby("city")["revenue_k"].sum()

# Do:
result = (
    df.query("country == 'BR'")
      .assign(revenue_k=lambda d: d.revenue / 1000)
      .groupby("city")["revenue_k"]
      .sum()
)

2. query() is faster to read than boolean masks

df.query("age > 30 and country in ['BR', 'PT']")

3. assign() for new columns inside a chain

df.assign(
    margin=lambda d: d.revenue - d.cost,
    margin_pct=lambda d: d.margin / d.revenue,
)

4. pipe() for custom steps

def winsorize(df, col, q=0.01):
    lo, hi = df[col].quantile([q, 1 - q])
    return df.assign(**{col: df[col].clip(lo, hi)})

df.pipe(winsorize, "revenue")

5–10: the rest

• Use .loc[] for assignment, never chained indexing.
• Categorical dtype saves memory and speeds up groupby.
• value_counts(normalize=True) gives you proportions in one call.
• merge() with indicator=True debugs join mismatches instantly.
• convert_dtypes() handles nullable ints and clean booleans.
• Profile with df.memory_usage(deep=True) before optimizing anything.

Adopt these and your future self (and your reviewers) will thank you.