matlab python linalg benchmark

matlab_bench.txt

>> disp('Eig');tic;data=rand(500,500);eig(data);toc;
disp('Svd');tic;data=rand(1000,1000);[u,s,v]=svd(data);s=svd(data);toc;
disp('Inv');tic;data=rand(1000,1000);result=inv(data);toc;
disp('Det');tic;data=rand(1000,1000);result=det(data);toc;
disp('Dot');tic;a=rand(1000,1000);b=inv(a);result=a*b-eye(1000);toc;
disp('Done');
Eig
Elapsed time is 0.840714 seconds.
Svd
Elapsed time is 1.931674 seconds.
Inv
Elapsed time is 0.478826 seconds.
Det
Elapsed time is 0.082375 seconds.
Dot
Elapsed time is 0.298401 seconds.
Done
>> disp('Eig');tic;data=rand(500,500);[a,b]=eig(data);toc;
disp('Svd');tic;data=rand(1000,1000);[u,s,v]=svd(data);s=svd(data);toc;
disp('Inv');tic;data=rand(1000,1000);result=inv(data);toc;
disp('Det');tic;data=rand(1000,1000);result=det(data);toc;
disp('Dot');tic;a=rand(1000,1000);b=inv(a);result=a*b-eye(1000);toc;
disp('Done');
Eig
Elapsed time is 0.511574 seconds.
Svd
Elapsed time is 1.913088 seconds.
Inv
Elapsed time is 0.181648 seconds.
Det
Elapsed time is 0.074878 seconds.
Dot
Elapsed time is 0.287743 seconds.
Done
>> disp('Eig');tic;data=rand(500,500);[a,b]=eig(data);toc;
disp('Svd');tic;data=rand(1000,1000);[u,s,v]=svd(data);s=svd(data);toc;
disp('Inv');tic;data=rand(1000,1000);result=inv(data);toc;
disp('Det');tic;data=rand(1000,1000);result=det(data);toc;
disp('Dot');tic;a=rand(1000,1000);b=inv(a);result=a*b-eye(1000);toc;
disp('Done');
Eig
Elapsed time is 0.548438 seconds.
Svd
Elapsed time is 2.069913 seconds.
Inv
Elapsed time is 0.176397 seconds.
Det
Elapsed time is 0.076114 seconds.
Dot
Elapsed time is 0.307366 seconds.
Done

python_bench.txt

In [3]: %timeit data=np.random.rand(500,500);a,b = np.linalg.eig(data)
1 loops, best of 3: 404 ms per loop

In [4]: %timeit data=np.random.rand(1000,1000);u,s,v = np.linalg.svd(data); snew=np.linalg.svd(data, full_matrices=False)
1 loops, best of 3: 2.11 s per loop

In [5]: %timeit data=np.random.rand(1000,1000);idata = np.linalg.inv(data);10 loops, best of 3: 198 ms per loop

In [6]: %timeit data=np.random.rand(1000,1000);ddata = np.linalg.det(data);
/Library/Frameworks/EPD64.framework/Versions/7.1/lib/python2.7/site-packages/numpy/linalg/linalg.py:1676: RuntimeWarning: overflow encountered in exp
  return sign * exp(logdet)
10 loops, best of 3: 68.6 ms per loop

In [7]: %timeit data=np.random.rand(1000,1000);ddata = np.linalg.det(data);
10 loops, best of 3: 68.5 ms per loop

In [8]: %timeit data=np.random.rand(1000,1000);idata = np.linalg.inv(data); result = np.dot(data,idata)-np.eye(1000)
1 loops, best of 3: 319 ms per loop

systeminfo.txt

 Model Name:	MacBook Pro
  Model Identifier:	MacBookPro5,2
  Processor Name:	Intel Core 2 Duo
  Processor Speed:	2.93 GHz
  Number of Processors:	1
  Total Number of Cores:	2
  L2 Cache:	6 MB
  Memory:	8 GB
  Bus Speed:	1.07 GHz
  Boot ROM Version:	MBP52.008E.B05
  SMC Version (system):	1.42f4



 System Version:	Mac OS X 10.7 (11A511)
  Kernel Version:	Darwin 11.0.0
  Boot Volume:	OSX
  Boot Mode:	Normal
  Secure Virtual Memory:	Enabled
  64-bit Kernel and Extensions:	Yes

What is the goal of this comparison? A few things puzzle me - why include random number generation in the timing? I'm assuming your interest is in the execution time of the the linear algebra routines. The other puzzle is why did you use random numbers? I'm guessing that performance of many of these algorithms could depend heavily on the structure of the data.

@scott: the goal was to essentially say that the performance of both were similar. as soon as i get to a power supply, i will update the benchmarks with taking the random number generator out of the loop and use the same data for both matlab and python. i simply don't have the time to show really specific comparisons. if you have data i can use, i'm happy to use the same for both. also, one additional purpose is to compare it to an equivalent boost routine.