[janzer]

And 80x is apparently actually an understatement, at least for a few cases. Some numbers recently posted to python-dev show up to a 124X improvement:

  Precision: 9 decimal digits

  float:
  result: 3.1415926535897927
  time: 0.113188s

  cdecimal:
  result: 3.14159265
  time: 0.158313s

  decimal:
  result: 3.14159265
  time: 18.671457s


  Precision: 19 decimal digits

  float:
  result: 3.1415926535897927
  time: 0.112874s

  cdecimal:
  result: 3.141592653589793236
  time: 0.348100s

  decimal:
  result: 3.141592653589793236
  time: 43.241220s

[Zenst]

Impressive stuff, though what struck me was the results, least accuracy wise and what rounding they using:

Pi is 3.14159 26535 89793 238....

So I do wonder what rounding they are using, even truncating as I have (next digit 4 so good place to do that) then can see the last digit should at least be 8, worst case 7 and 6!! There again this may be a convention or result of the methord to calulate Pi.

As for floats, well, for accuracy I'd go with cdecimal right there, though is it as accurate. I suspect it is the formular used that induces the minute error in results.

http://en.wikipedia.org/wiki/Pi #21 reference

[janzer]

This is from the decimal benchmarks included in the python source[1], in the recipe given in the decimal documentation[2] the precision is increased for the intermediate steps of the algorithm so it gives the correct end result.

  >>> pi()
  Decimal('3.141592653589793238')

1. http://hg.python.org/cpython/file/344d67063c8f/Modules/_deci...

2. http://docs.python.org/library/decimal.html#recipes

[Zenst]

Thank you, out of interest on a appliction I work on every now and then showed a 50% improvement with the code as is, not heavy decimal at all, mostly int's though still a nice speedup.