Re: puzzling difference in speed
- To: mathgroup at smc.vnet.net
- Subject: [mg34877] Re: puzzling difference in speed
- From: "Carl K. Woll" <carlw at u.washington.edu>
- Date: Tue, 11 Jun 2002 05:00:46 -0400 (EDT)
- Organization: University of Washington
- References: <200206040741.DAA03886@smc.vnet.net> <adpfh5$j45$1@smc.vnet.net>
- Sender: owner-wri-mathgroup at wolfram.com
Hi all,
Sorry if this is a repost, my previous reply seems to have been lost.
The reason for the difference in speed is the use of packed arrays.
Repeating Fred Simon's form of the original example:
In[22]:=
n=50;t1=Table[0,{n},{n},{n}];
In[23]:=
t=array[Plus,Dimensions[t1]];
u=array[Plus,{n,n,n}];
t===u
Out[25]=
True
So, as far as we can tell, t and u are the same. But, when we apply Array to
each of the expressions we get:
In[26]:=
Apply[Array,t];//Timing
Apply[Array,u];//Timing
Out[26]=
{0.14 Second, Null}
Out[27]=
{0.016 Second, Null}
So, t and u must not be the same, since the timings are so different. Let's
see if PackedArrays are being used:
In[28]:=
Developer`PackedArrayForm[t]
Developer`PackedArrayForm[u]
Out[28]=
array[Plus, PackedArray[Integer, <3>]]
Out[29]=
array[Plus, {50, 50, 50}]
Now we see the culprit, t and u aren't really identical, as t uses
PackedArrays and u doesn't. In this particular case, having packed arrays is
bad, so try instead:
In[32]:=
Apply[Array,Developer`FromPackedArray[t]];//Timing
Apply[Array,u];//Timing
Out[32]=
{0.031 Second, Null}
Out[33]=
{0.016 Second, Null}
You still might be puzzled as to why the packed array case is so slow. The
answer is a bit paradoxical. In the case where the dimensions of the array
are strictly integers, Mathematica looks at the size of the matrix to be
produced. If the matrix is sufficiently large, and if only machine numbers
are involved, then Mathematica's packed array technology kicks in and the
answer is produced very quickly. On the other hand, if a packed array is
present, that is the dimensions of the array is a packed array, then
Mathematica doesn't use the packed array technology to compute the result.
Consider
In[61]:=
Apply[Array,t]//Developer`PackedArrayQ
Out[61]=
False
In[62]:=
Apply[Array,u]//Developer`PackedArrayQ
Out[62]=
True
So, the moral of the story is that packed arrays can effect your results in
surprising ways.
Carl Woll
Physics Dept
U of Washington
"Fred Simons" <f.h.simons at tue.nl> wrote in message
news:adpfh5$j45$1 at smc.vnet.net...
> Hartmut Wolf remarked with respect to the given examples::
>
> > Obviously the computing machinery for Array behaves differently when the
> > dimensions are given explicitly or introduced as an expression (to be
> > evaluated)
> >
>
>
> It seems to be more complicated. Have a look at the following results:
>
> In[1]:=
> n=100; t1 = Table[0, {n},{n},{n}];
>
> In[2]:=
> t = array[Plus, Dimensions[t1]];
> u = array[Plus, {100, 100, 100}];
> Equal[t, u]
>
> Out[4]= True
>
> In[5]:=
> ReplaceAll[t, array\[Rule]Array]; // Timing
> ReplaceAll[u ,array\[Rule]Array]; // Timing
>
> Out[5]=
> {0.65 Second,Null}
> Out[6]=
> {0.77 Second,Null}
>
> In[7]:=
> Apply[Array, t]; // Timing
> Apply[Array, u]; // Timing
>
> Out[7]=
> {4.51 Second,Null}
> Out[8]=
> {0.72 Second,Null}
>
> Despite the fact that t equals u, we have the same difference in timing.
> Does Mathematica 'remember' the way the expression t has been computed?
>
> Fred Simons
> Eindhoven University of Technology
>
>
>
- References:
- RE: puzzling difference in speed
- From: "Wolf, Hartmut" <Hartmut.Wolf@t-systems.com>
- RE: puzzling difference in speed