[Numpy-discussion] Permutations in Simulations`
Keith Goodman
kwgoodman@gmail....
Tue Feb 10 20:06:58 CST 2009
Yeah, good point. The second argsort isn't needed. That should speed things up.
The double argsort ranks the values in the array. But we don't need that here.
On Tue, Feb 10, 2009 at 5:31 PM, <josef.pktd@gmail.com> wrote:
> very nice. What's the purpose of the second `.argsort(0)` ? Doesn't
> it also work without it, or am I missing something in how this works?>
>
> Josef
>
> On 2/10/09, Mark Janikas <mjanikas@esri.com> wrote:
>> Thanks to all for your replies. I want this to work on any vector so I was
>> thinking this...?
>>
>> import numpy as np
>> import timeit
>> x = np.array([4.,5.,10.,3.,5.,6.,7.,2.,9.,1.])
>> nx = 10
>> ny = 100
>>
>> def weirdshuffle4(x, ny):
>> nx = len(x)
>> indices = np.random.random_sample((nx,ny)).argsort(0).argsort(0)
>> return x[indices]
>>
>> t=timeit.Timer("weirdshuffle4(x,ny)", "from __main__ import *")
>> print t.timeit(100)
>>
>> 0.0148663153873
>>
>>
>> -----Original Message-----
>> From: numpy-discussion-bounces@scipy.org
>> [mailto:numpy-discussion-bounces@scipy.org] On Behalf Of Keith Goodman
>> Sent: Tuesday, February 10, 2009 12:59 PM
>> To: Discussion of Numerical Python
>> Subject: Re: [Numpy-discussion] Permutations in Simulations`
>>
>> On Tue, Feb 10, 2009 at 12:41 PM, Keith Goodman <kwgoodman@gmail.com> wrote:
>>> On Tue, Feb 10, 2009 at 12:28 PM, Keith Goodman <kwgoodman@gmail.com>
>>> wrote:
>>>> On Tue, Feb 10, 2009 at 12:18 PM, Keith Goodman <kwgoodman@gmail.com>
>>>> wrote:
>>>>> On Tue, Feb 10, 2009 at 11:29 AM, Mark Janikas <mjanikas@esri.com>
>>>>> wrote:
>>>>>> I want to create an array that contains a column of permutations for
>>>>>> each
>>>>>> simulation:
>>>>>>
>>>>>> import numpy as NUM
>>>>>>
>>>>>> import numpy.random as RAND
>>>>>>
>>>>>> x = NUM.arange(4.)
>>>>>>
>>>>>> res = NUM.zeros((4,100))
>>>>>>
>>>>>>
>>>>>> for sim in range(100):
>>>>>>
>>>>>> res[:,sim] = RAND.permutation(x)
>>>>>>
>>>>>>
>>>>>> Is there a way to do this without a loop? Thanks so much ahead of
>>>>>> time.
>>>>>
>>>>> Does this work? Might not be faster but it does avoid the loop.
>>>>>
>>>>> import numpy as np
>>>>>
>>>>> def weirdshuffle(nx, ny):
>>>>> x = np.ones((nx,ny)).cumsum(0, dtype=np.int) - 1
>>>>> yidx = np.ones((nx,ny)).cumsum(1, dtype=np.int) - 1
>>>>> xidx = np.random.rand(nx,ny).argsort(0).argsort(0)
>>>>> return x[xidx, yidx]
>>>>
>>>> Hey, it is faster for nx=4, ny=100
>>>>
>>>> def baseshuffle(nx, ny):
>>>> x = np.arange(nx)
>>>> res = np.zeros((nx,ny))
>>>> for sim in range(ny):
>>>> res[:,sim] = np.random.permutation(x)
>>>> return res
>>>>
>>>>>> timeit baseshuffle(4,100)
>>>> 1000 loops, best of 3: 1.11 ms per loop
>>>>>> timeit weirdshuffle(4,100)
>>>> 10000 loops, best of 3: 127 µs per loop
>>>>
>>>> OK, who can cut that time in half? My first try looks clunky.
>>>
>>> This is a little faster:
>>>
>>> def weirdshuffle2(nx, ny):
>>> one = np.ones((nx,ny), dtype=np.int)
>>> x = one.cumsum(0)
>>> x -= 1
>>> yidx = one.cumsum(1)
>>> yidx -= 1
>>> xidx = np.random.random_sample((nx,ny)).argsort(0).argsort(0)
>>> return x[xidx, yidx]
>>>
>>>>> timeit weirdshuffle(4,100)
>>> 10000 loops, best of 3: 129 µs per loop
>>>>> timeit weirdshuffle2(4,100)
>>> 10000 loops, best of 3: 106 µs per loop
>>
>> Sorry for all the mail.
>>
>> def weirdshuffle3(nx, ny):
>> return np.random.random_sample((nx,ny)).argsort(0).argsort(0)
>>
>>>> timeit weirdshuffle(4,100)
>> 10000 loops, best of 3: 128 µs per loop
>>>> timeit weirdshuffle3(4,100)
>> 10000 loops, best of 3: 37.5 µs per loop
>> _______________________________________________
>> Numpy-discussion mailing list
>> Numpy-discussion@scipy.org
>> http://projects.scipy.org/mailman/listinfo/numpy-discussion
>>
>> _______________________________________________
>> Numpy-discussion mailing list
>> Numpy-discussion@scipy.org
>> http://projects.scipy.org/mailman/listinfo/numpy-discussion
>>
> _______________________________________________
> Numpy-discussion mailing list
> Numpy-discussion@scipy.org
> http://projects.scipy.org/mailman/listinfo/numpy-discussion
>
More information about the Numpy-discussion
mailing list