MathGroup Archive 2000

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Unusual speedup of Table

  • To: mathgroup at smc.vnet.net
  • Subject: [mg25617] Re: [mg25605] Unusual speedup of Table
  • From: Daniel Lichtblau <danl at wolfram.com>
  • Date: Mon, 16 Oct 2000 03:04:40 -0400 (EDT)
  • References: <200010110750.DAA18883@smc.vnet.net>
  • Sender: owner-wri-mathgroup at wolfram.com

SATether wrote:
> 
> Why is the second call of Table below so much faster than the first? These
> results are from Mathematica 4.0.2.1 for Power Mac (400 MHz PPC 750, OS 9). It
> isn't some kind of cache effect because the times are consistent over many
> executions of both expressions. (And yes, I know I can do a lot better still
> for this particular computation with Drop and Take).
> 
> - Steve Tether
> 
> n = 10000;
> a = Table[Random[], {n}];
> Timing[Table[a[[j + 1]] - 2 a[[j]] + a[[j - 1]], {j, 2, n - 1}];]
> 
> {10.5333 Second, Null}
> 
> Timing[Table[Switch[j, _, a[[j + 1]] - 2 a[[j]] + a[[j - 1]]], {j, 2, n - 1}];]
> 
> {0.7 Second, Null}

I do not know why the first Table construction is slow. I suspected some
needless unpacking, but this is not the case. Specifically, if I first
do

Developer`SetSystemOptions["UnpackMessage"->True];

then I get no message when running your Table construction.

On a more optimistic note, this slowness has apparently been addressed
in the Mathematica kernel that will be in a future release. Below I show
this on a list 100 x as large as yours.

In[6]:= n = 1000000;

In[7]:= a = Table[Random[], {n}];

In[8]:= Timing[t1 = Table[a[[j + 1]] - 2 a[[j]] + a[[j - 1]], {j, 2, n -
1}];]
Out[8]= {25.17 Second, Null}

This is significantly faster than what I get using the Table[Switch...]]
workaround, as it should be.

Note that one can significantly improve speed of such a computation
using ListCorrelate.

In[9]:= Timing[t2 = ListCorrelate[{1,-2,1}, a];]

Out[9]= {1.7 Second, Null}

In[10]:= Max[Abs[t1-t2]]
                   -16
Out[10]= 4.44089 10


Daniel Lichtblau
Wolfram Research


  • Prev by Date: A function to evaluate only parts matching a pattern
  • Next by Date: Re: tracing lines on surface-graphics
  • Previous by thread: Unusual speedup of Table
  • Next by thread: Re: Unusual speedup of Table