Single Value Decomposition function for dojo.math.matrix

[enzo]

The thinsvd(A) function in the attached SVD.js implements the SVD for a matrix A built as 2-D array of arrays, and can easily be added to dojo.math.matrix as it does not have external dependencies. The code is derived from the Java implementation in JAMA (​http://math.nist.gov/javanumerics/jama/ ), which is in the public domain. The attached SVDtest.html is a self-contained HTML+JS demo page.

With square random matrices, complexity grows roughly as O(n³). Decomposing a 100x100 matrix typically takes, with a single-core 1.7GHz Pentium M:

• Chrome 3.0.195.27: 880 ms
• Firefox 3.5.4: 950 ms
• Safari 4.0.3: 1,360 ms
• MSIE 8.0: 9,554 ms (and three "slow script" warnings...)

For further details, see the comments in SVD.js .

[enzo]

SVD code

[enzo]

HTML+JS demo page (also shows use of SVD to calculate (Pseudo-)Inverse

[Tom Trenka]

Given the age of these tickets and the small nature of them, I'm marking all as won't fix. #1395 was somewhat addressed with the implementation of SHA2 in DTK 1.8.