Python float

Created with Sketch.

Python float

Summary: in this tutorial, you’ll learn about the Python float type, how Python represents floating-point numbers, and how to test the floating-point number for equality.

Introduction to the Python float type

Python uses the float class to represent real numbers.

CPython implements float using C double type. The C double type usually implements IEEE 754 double-precision binary float, which is also called binary64.

Python float uses 8 bytes (or 64 bits) to represent real numbers. Unlike the integer type, the float type uses a fixed number of bytes.

Technically, Python uses 64 bits as follows:

  • 1 bit for sign (positive or negative)
  • 11 bits for exponent 1.5e-5 1.5 x 10-5 (exponent is -5) the range is [-1022, 1023].
  • 52 bits for significant digits

For the sake of simplicity, significant digits are all digits except leading and trailing zeros.

For example, 0.25 has two significant digits, 0.125 has three significant digits, and 12.25 has four significant digits.

(1.25)10 = (1×20 + 0x2-1 + 1×2-2)10 = (1.01)2

Some numbers have a finite binary representation, but some don’t, e.g., 0.1. It’s 01.0001100110011... in binary.

Because of this, Python can only use approximate float representations for those numbers.

Python float class

The float() returns a floating-point number based on a number or a string. For example:

>>> float(0.1)
>>> float('1.25')

Code language: JavaScript (javascript)

If you pass an object (obj) to the float(obj), it’ll delegate to the obj.__float__(). If the __float__() is not defined, it’ll fall back to __index__().

If you don’t pass any argument to the float(), it’ll return 0.0

When you use the print() function, you’ll see that the number 0.1 is represented as 0.1 exactly.

Internally, Python can only represent 0.1 approximately.

To see how Python represents the 0.1 internally, you can use the format() function.

The following shows how Python represents the number 0.1 using 20 digits:

>>> format(0.1, '.20f')

Code language: JavaScript (javascript)

As you can see, 0.1 is not exactly 0.1 but 0.10000000000000000555...

Because Python can represent some floats approximately, it will cause many problems when you compare two floating-point numbers.

Equality testing

Let’s take a look at the following example:

x = 0.1 + 0.1 + 0.1
y = 0.3

print(x == y)

Code language: PHP (php)



Code language: PHP (php)

Internally, Python cannot use a finite number of digits to represent the numbers x and y:

print(format(x, '.20f'))
print(format(y, '.20f'))

Code language: PHP (php)



Code language: CSS (css)

Note that the number of digits is infinite. We just show the first 20 digits.

One way to work around this problem is to round both sides of the equality expression to a number of significant digits. For example:

x = 0.1 + 0.1 + 0.1
y = 0.3
print(round(x, 3) == round(y, 3))

Code language: PHP (php)



Code language: PHP (php)

This workaround doesn’t work in all cases.

PEP485 provides a solution that fixes this problem by using relative and absolute tolerances.

It provides the isclose() function from the math module returns True if two numbers are relatively close to each other.

The following shows the isclose() function signature:

isclose(a, b, rel_tol=1e-9, abs_tol=0.0)


For example:

from math import isclose

x = 0.1 + 0.1 + 0.1
y = 0.3


Code language: JavaScript (javascript)



Code language: PHP (php)


  • Python uses float class to represent real numbers.
  • Python uses a fixed number of bytes (8 bytes) to represent floats. Therefore, it can represent some numbers in binary approximately.
  • Use the isclose() function from the math module to test equality for floating-point numbers.

Leave a Reply

Your email address will not be published. Required fields are marked *