Services & Resources / Wolfram Forums
-----
 /
MathGroup Archive
1995
*January
*February
*March
*April
*May
*June
*July
*August
*September
*October
*November
*December
*Archive Index
*Ask about this page
*Print this page
*Give us feedback
*Sign up for the Wolfram Insider

MathGroup Archive 1995

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

Search the Archive

Speeding Up Numerical Matrix Calculations ???

  • To: mathgroup at christensen.cybernetics.net
  • Subject: [mg1091] Speeding Up Numerical Matrix Calculations ???
  • From: Paul E Howland <PEHOWLAND at taz.dra.hmg.gb>
  • Date: Sun, 14 May 1995 21:11:31 -0400

It is common in many signal processing applications to formulate problems 
in terms of simple functions of lots of matrices.  Hence, I guess, the 
existence of MATLAB.

My question is quite simple.  Mathematica can be numerically very slow.  
How can I speed up my numerical calculations involving matrices that 
consist only of machine size numbers?

It appears that Mathematica handles numerical matrices in much the same
way as it does symbolic matrices.  Here's a test I did using Mathematica
V2.2.3 on a 486/33 PC running Windows 3.1. 

In[1]:= F[a_?MatrixQ, b_?MatrixQ, c_?MatrixQ] :=
	Transpose[a]a + 3 b c + Transpose[c]c
In[2]:= NF[a_?MatrixQ, b_?MatrixQ, c_?MatrixQ] :=
	N[Transpose[a]a + 3 b c + Transpose[c]c]

Thus I've defined two (arbitrary) functions that manipulate matrices.
NF is simply N[F] - I want to see whether wrapping N[] around a calculation
causes Mathematica to invoke some numerically optimised code.

Now generate some machine sized numerical matrices ...

In[3]:= a = Table[Random[], {i,10}, {j,10}];
In[4]:= b = Table[Random[], {i,10}, {j,10}];
In[5]:= c = Table[Random[], {i,10}, {j,10}];

And some symbolic matrices ...

In[6]:= x = Table[FromCharacterCode[Random[Integer, {100,122}]],
            {i, 10}, {j,10}];
In[7]:= y = Table[FromCharacterCode[Random[Integer, {100,122}]],
            {i, 10}, {j,10}];
In[8]:= z = Table[FromCharacterCode[Random[Integer, {100,122}]],
            {i, 10}, {j,10}];

And compare timings.  First compare F[] on both sets of data ...

In[9] := Timing[Do[F[a,b,c], {i,10}]]
Out[9] = {3.185 Second, Null}
In[10] := Timing[Do[F[x,y,z], {i,10}]]
Out[10] = {2.087 Second, Null}

thus the symbolic processing is faster (!).  Now see what the effect
of wrapping N[] around the code is, for numerical data...

In[11] := Timing[Do[NF[a,b,c], {i,10}]]
Out[11] = {3.24 Second, Null}

... it makes it marginally slower!  Presumably due to the extra time 
required by Mathematica to interpret the N[] function.

It is not possible to use Compile[] on a function that accepts matrices 
as arguments because Compile[] only accepts _Integer, _Real and _Complex 
as its data types (and not _List).

Is there any means of getting Mathematica to manipulate numerical matrices
more quickly (other than using Matlab or writing a C program and then
using MathLink)?  It would be nice if Mathematica was written such that
when N[] was wrapped around simple numerical matrix operations, optimised
numerical code was invoked. 

Paul E Howland                         

Long Range Ground Radar Systems Section               tel. +44 (0)1684 895767
LSC2 Division, Room BY209                             fax. +44 (0)1684 896315
Defence Research Agency                           email: PEHOWLAND at DRA.HMG.GB 
Malvern, Worcs, WR14 3PS, UK.                      
-----------------------------------------------------------------------------



  • Prev by Date: Re: How to solve system of inequalities?
  • Next by Date: Re: Pasting Mathematica notebooks into Word
  • Previous by thread: Re: Online documentation available?
  • Next by thread: Re: Speeding Up Numerical Matrix Calculations ???