[ojAlgo-user] Performance tests

[Anders P. <ap...@op...>]

I regularly test performance and make comparisons between ojAlgo and 
Jama (also part of ojAlgo).

The MatrixDecomposition interface has two major methods: compute() and 
solve().

Here are example figures using 1000x1000 matrices and performing each 
operation 10 times.

compute() ojAlgo/Jama

Chol: 8,15/7,09
LU: 20,39/14,56
QR: 98,68/104,51

solve() ojAlgo/Jama

Chol: 39,88/197,55
LU: 41,28/61,69
QR: 177,58/268,45


Comments
--------------

Don't use jama cholesky solve, it's very (unnecessarily) slow!

Apart from the cholesky case, solve takes about 2 to 4 times longer than 
compute.

Jama performs better with compute, but ojAlgo does solve faster. Doing 
both compute and solve (which is often what you want) ojAlgo is 20-25% 
faster than Jama (77% in the Cholesky case).

ojAlgo, and QR in particular, is favored by large matrices when 
comparing to Jama. The two QR implementations are about equally fast 
when using 400x400 matrices.

For SVD and Eigenvalue decompositions ojAlgo delegates to Jama - there 
is nothing to compare.

The source for these tests are in CVS.

You can see more of the results here:
http://ojalgo.org/generated/comparison.xml


Don't forget!!! When using ojAlgo you can easily switch between using 
the different implementations.


/Anders