lemire / fastrand Goto Github PK

Hi,

I'm a little novice, who would like to solve the following little problem:

1. Say I can control when to generate a number using fastrand (bounded for instance by 1001).
2. How can I guess at best its interval?

I mean:

1. suppose now, fastrand generate the number 989.
2. is there a way to figure out a lower/upper bound, such as 900 to 1000, please?

Regards,

it seems that the numbers that are generated are the same every time a program is run.

Following this issue...
#9

We should now be able to do "pip install fastrand" on Linux and Windows. However, macOS remains unsupported.

Is there a way to set our own seed for reproducibility purposes? Similarly for ensuring different streams while parallelizing code...

When installing for Python3.5, at import time it gives an error:

undefined symbol: PyInt_AsUnsignedLongLongMask

to solve it, I had to modify the .c file from:

#if PY_MAJOR_VERSION >= 3
#define PyInt_AsLong(x) PyLong_AsLong(x)
#endif

to

#if PY_MAJOR_VERSION >= 3
#define PyInt_AsLong(x) PyLong_AsLong(x)
#define PyInt_AsUnsignedLongLongMask(x) PyLong_AsUnsignedLongLongMask(x)
#endif

It would be great if you could upload the package so that it can be installed via pip install fastrand. The current process is a bit tedious.

Add an option to specify a lower bound for the random int generation.
Examples from numpy and the random lib in python below.

numpy.random.randint(950, 1000)
random.randint(100,1000)

This would make fastrand a better replacement for quickly generating random ints.

I peek at the code pcg32_random_bounded_divisionless().
Why variable 'leftover' able to reject unbiased randoms ?

No bias if leftover >= range or leftover(updated) >= threshold.
I don't follow above logic ...

Does this produce unbiased ranged number ?
Or, is it an approximation, using your idea in another blog:

https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/

Error:

Collecting fastrand
  Downloading https://files.pythonhosted.org/packages/4a/c9/84ae3fa10cb24bd5ac88845a55ddddc572d165218e29613063763f12d8c4/fastrand-1.7.0.tar.gz
    Complete output from command python setup.py egg_info:
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/tmp/pip-build-lr57d0c2/fastrand/setup.py", line 9, in <module>
        long_description=open('README.md', 'r').read(),
      File "/usr/lib/python3.5/encodings/ascii.py", line 26, in decode
        return codecs.ascii_decode(input, self.errors)[0]
    UnicodeDecodeError: 'ascii' codec can't decode byte 0xe2 in position 124: ordinal not in range(128)
    
    ----------------------------------------
Command "python setup.py egg_info" failed with error code 1 in /tmp/pip-build-lr57d0c2/fastrand/

Any way to fix this?

I know this is not officially supported but is there any way to install this library on Windows?

The C function pcg32_random() returns an unsigned uint32_t, but the pcg32() function exposed to Python calls Py_BuildValue() with "i", which causes it to return a signed int (docs). Is this a bug or was it intentional? I don't see this explicitly mentioned anywhere, but I assumed it would give me unsigned 32-bit integers, and I was surprised to see it giving me negative numbers.

On latest python 2.7 your library is faster

python -m timeit -s 'import fastrand' 'fastrand.pcg32bounded(1001)'
10000000 loops, best of 3: 0.0994 usec per loop

python -m timeit -s 'import numpy' 'numpy.random.randint(0, 1000)'
1000000 loops, best of 3: 1.07 usec per loop

so, I think this should be removed from readme :)

David Andersen points out that
python -m timeit -s 'import numpy' 'numpy.random.randint(0, 1000)'
is much faster.

Hi to all,

We tried to download latest version of fastrand 1.2.
However "pip install fastrand" ended up installing 1.1
"pip install fastrand --upgrade" also ended up installing 1.1
What can be the reason for that?

Best,
Burcak Otlu

Subj.

?

Thank you for your really fast random number generator.

;)

To reproduce:

python -m timeit -s 'import fastrand' 'fastrand.pcg32_seed(13)'

Fatal Python error: deallocating None

Current thread 0x00000001183f3dc0 (most recent call first):
  File "<timeit-src>", line 6 in inner
  File "/usr/local/Cellar/python/3.7.6_1/Frameworks/Python.framework/Versions/3.7/lib/python3.7/timeit.py", line 176 in timeit
  File "/usr/local/Cellar/python/3.7.6_1/Frameworks/Python.framework/Versions/3.7/lib/python3.7/timeit.py", line 222 in autorange
  File "/usr/local/Cellar/python/3.7.6_1/Frameworks/Python.framework/Versions/3.7/lib/python3.7/timeit.py", line 324 in main
  File "/usr/local/Cellar/python/3.7.6_1/Frameworks/Python.framework/Versions/3.7/lib/python3.7/timeit.py", line 374 in <module>
  File "/usr/local/Cellar/python/3.7.6_1/Frameworks/Python.framework/Versions/3.7/lib/python3.7/runpy.py", line 85 in _run_code
  File "/usr/local/Cellar/python/3.7.6_1/Frameworks/Python.framework/Versions/3.7/lib/python3.7/runpy.py", line 193 in _run_module_as_main
zsh: abort      python -m timeit -s 'import fastrand' 'fastrand.pcg32_seed(13)'

Or from within python:

import fastrand; rngs = [fastrand.pcg32_seed(i) for i in range(100000)]

Fatal Python error: deallocating None

Current thread 0x0000000116acadc0 (most recent call first):
  File "<stdin>", line 1 in <module>
zsh: abort      python

Verified that it happens on multiple OSes (MacOS, ScientificLinux).

To match pcg32bounded() doc-string (below), I propose to patch argument check.
Sorry, this should be a pull-request, but I don't know how to use it

generate random integer in the interval [0,range) using PCG.

Since pcg32bounded only allowed 1 argument, the code can be optimized
without overhead of parsing tuple. Just go straight for the number.

This is the full patch:

56,58c56,58
<      int n;
<     if (!PyArg_ParseTuple(args, "i", &n))
<         return NULL;
---
>     int n = PyLong_AsLong(args);   // 1 argument, ok for Python 2 or 3
>     if (n <= 0) return PyErr_Occurred() ? NULL : (Py_INCREF(Py_None), Py_None);
>
98c98
<      {"pcg32bounded", pcg32bounded, METH_VARARGS, "generate random integer in
the interval [0,range) using PCG."},
---
>      {"pcg32bounded", pcg32bounded, METH_O, "generate random integer in the in
terval [0,range) using PCG."},

With the patch, this is benchmark on my laptop:

> python -m timeit -s "import fastrand" "fastrand.pcg32bounded(1001)"
1000000 loops, best of 3: 0.208 usec per loop   -- original
10000000 loops, best of 3: 0.144 usec per loop  -- patched

Error messages now as expected:

>>> from fastrand import *
>>> print pcg32bounded(1001)
82
>>> print pcg32bounded(0)       
None                          # should not return anything
>>> print pcg32bounded(2**31-1), pcg32bounded(-2**31)
20852737 None                 # max input limits
>>> print pcg32bounded(2**31) # overflow 
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OverflowError: long int too large to convert to int
>>> print pcg32bounded('Hi')  # bad integer
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: an integer is required
>>>