Decimal Python vs. float runtime

Question

Just a general question on what sort of runtime differences I should be expecting between using these two different data types.

My test:

test = [100.0897463, 1.099999939393,1.37382829829393,29.1937462874847272,2.095478262874647474]
test2 = [decimal.Decimal('100.0897463'), decimal.Decimal('1.09999993939'), decimal.Decimal('1.37382829829'), decimal.Decimal('29.1937462875'), decimal.Decimal('2.09547826287')]

def average(numbers, ddof=0):
    return sum(numbers) / (len(numbers)-ddof)

%timeit average(test)
%timeit average(test2)

The differences in runtime are:
1000000 loops, best of 3: 364 ns per loop
10000 loops, best of 3: 80.3 µs per loop

So using decimal was about 200 times slower than using floats. Is this type of difference normal and along the lines of what I should expect when deciding which data type to use?

Answer 1

You get better speed with float because Python float uses the hardware floating point register when available (and it is available on modern computers), whereas Decimal uses full scalar/software implementation.

However, you get better control with Decimal, when you have the classical floating point precision problems with the float types. See the classical StackOverflow Q&A Is floating point math broken? for instance.

Answer 2

Based on the time difference you are seeing, you are likely using Python 2.x. In Python 2.x, the decimal module is written in Python and is rather slow. Beginning with Python 3.2, the decimal module was rewritten is C and is much faster.

Using Python 2.7 on my system, the decimal module is ~180x slower. Using Python 3.5, the decimal module is in only ~2.5x slower.

If you care about decimal performance, Python 3 is much faster.