DataTerminology.comNEW
Also known as: Floating-Point
A numeric data type representing real numbers with a fractional component using finite binary approximation, commonly defined by IEEE 754 formats.
DESCRIPTION
Float (floating-point) represents real numbers in binary with separate sign, exponent, and mantissa fields, enabling wide dynamic range but limited precision. Common formats include 32-bit single precision and 64-bit double precision. Operations can introduce rounding error, cancellation, and comparison pitfalls, with special values for NaN and infinity. For currency or exact arithmetic, decimal or fixed-point types are preferred. Careful tolerance-based comparisons and formatting avoid misleading results.
ALTERNATIVE DEFINITIONS
Python Documentation
Floating point numbers are represented in computer hardware as base 2 (binary) fractions.
Highlights binary representation cause of precision issues.
Oracle Java Tutorials
float: The float data type is a single-precision 32-bit IEEE 754 floating point.
Defines float size and IEEE 754 conformance in Java.
PostgreSQL Documentation
The data types real and double precision are inexact, variable-precision numeric types.
Notes inexactness for database float types and precision trade-offs.
RELATED TERMS
Double Precision Single Precision IEEE 754 NaN Infinity Rounding Error Precision Numeric Type
COMPETING TERMS
KNOWLEDGE AREAS
Programming Languages Software Engineering Data Modeling Data Storage & Operations
TARGET AUDIENCE
Software Developer Data Engineer Data Scientist Analytics Engineer
EXTERNAL SOURCES
- Wikipedia - Floating-point arithmetic
- IEEE - IEEE 754 floating-point arithmetic standard
- Python Documentation - Python guide to floating-point issues
- cppreference - C++ floating-point type reference
TOP QUESTIONS
Why do Float calculations sometimes produce surprising results?
Binary representation can’t exactly encode many decimal fractions, causing rounding errors. Use tolerances for comparisons and decimal types for financial calculations.
Category: Misunderstanding
When should you use Decimal instead of Float?
For exact arithmetic such as currency, Decimal avoids rounding surprises. Floats are better for scientific computing requiring wide dynamic range.
Category: Best practices
What are NaN and Infinity in Float arithmetic?
IEEE 754 defines special values: NaN for undefined results and ±Infinity for overflow or division by zero. Handle them explicitly in validations.
Category: Requirements
How do you compare Float values safely?
Use an epsilon tolerance or relative comparison. Avoid direct equality checks because tiny rounding differences may cause false mismatches.
Category: Testing
What is catastrophic cancellation?
Subtracting nearly equal floats loses significant precision. Reorder computations or use numerically stable algorithms to mitigate.
Category: Risk
Does Float precision differ across languages?
Most languages follow IEEE 754, but type names and default precisions differ (e.g., Python float is double precision, Java float is 32-bit).
Category: Requirements
How do you serialize Float safely?
Use standardized formats and precision control, avoid JSON scientific notation pitfalls, and consider decimal serialization when exactness matters.
Category: Best practices
What’s the difference between Float and Double?
Double typically uses 64 bits for greater precision and range than 32-bit float, at the cost of memory and sometimes performance.
Category: Example
Can Float overflow or underflow occur?
Yes. Overflow yields Infinity; underflow produces subnormal numbers or zero. Monitor ranges and rescale inputs if necessary.
Category: Risk