The files in this folder represent the profiling results for fastnumbers functionality. The results are separated by Python version. You can see what is being profiled in profile.py.

Some general observations about the results:

• The older the Python version, the greater the benefit fastnumbers brings over the built-in functionality. This is because the CPython maintainers are continiously making Python faster.
• The conversion to floats is always faster than built-in functionality for Python <= 3.9. Starting with Python 3.10 some optimization was added that makes the conversion for small floats very fast.
• The conversion to integers is always faster than built-in functionality until the length of the input digit string is greater than 18 characters (the cut-off for 64-bit integers), at which point it is roughty tied to slightly slower than CPython.
• When numbers are provided (as opposed to strings), the Python built-ins are usually faster.
• If you need to convert a list of data, using the map option is 2x faster than using try_* in a list comprehension, and about 1.5x faster than using try_* with the builtin map function.
  • Returning an iterator and returning a list are quite similar in performance.
• If you need to convert a list of data to a numpy array, try_array is about 2x faster than using the map option and then converting the resulting list to an ndarray. Interestingly, try_array is only slightly faster than using the map option by itself.