4th Edition Algorithm: Singular Value Decomposition

[norvig]

We will implement SVD, see for example http://stats.stackexchange.com/questions/159325/what-fast-algorithms-exist-for-computing-truncated-svd

[kanishk1225]

Is this implemented in java?

[norvig]

we would want implementations in java and python, and perhaps a demo in javascript

[sofmonk]

Sir, so we first start work on writing the pseudocode?

[Chipe1]

@norvig Are we supposed to follow the method given by amoeba (top answer on the stackexchange question)? I tried doing A = [0 B;B' 0] but instead of converging I encounter a loop. I couldn't find a proof for equivalence of truncated SVD of B and eigen decomposition of [0 B;B' 0]. I was able to prove it for [B 0; 0 B'] instead.
Also, are we finding only the biggest eigen value? If not, do I generate a new X using Gram–Schmidt?
I'm not sure how/where we are going to use SVD. My guess is dimensionality reduction, so we could need more eigenvalues/vectors.
Is this a step in the right direction?

My bad. Just had to normalize the parts individually instead of the whole vector X.

[Shashi456]

Can i work on this issue ?

[Chipe1]

@Shashi456 I've already implemented it in python. You can improve it, write a pseudocode, or implement it in another language.

[Shashi456]

Could i take a look at the code for writing the pseudocode then ?

[Chipe1]

Link to code
The code is not very clear and does some things to prevent precision problems. The pseudocode won't require doing that.

[Shashi456]

Thanks @Chipe1 I'll take a look at it , have you applied just some algorithm generically or did you take a look at existing algo's and converted them into code?