Dtypes and casting

Each column has a dtype controlling memory and valid operations. Cast explicitly with astype, pd.to_numeric, or pd.to_datetime—especially after CSV loads.

Common dtypes

• int64, float64 — numeric
• object — Python strings or mixed (often strings from CSV)
• bool — True/False
• category — low-cardinality strings (memory efficient)
• datetime64[ns] — timestamps
• Int64 (capital I) — nullable integer with NaN support

Safe conversion

import pandas as pd
s = pd.Series(['1', '2', 'bad', '4'])
nums = pd.to_numeric(s, errors='coerce')  # 'bad' → NaN
print(nums)

Downcasting

Use smallest dtype that fits (pd.to_numeric(..., downcast='integer')) on large datasets to save memory. Always validate after casting.

Important interview questions and answers

1. Q: object dtype?
   A: Often means strings from CSV—convert before math operations.
2. Q: errors='coerce'?
   A: Invalid values become NaN instead of raising—good for messy real data.

Self-check

1. Convert a string column to float with bad values coerced.
2. What dtype supports nullable integers?

Pitfall: astype(int) on floats truncates—round or use errors='coerce' first.