Segfault on large SVDs

[ryana]

Is there a practical limit on the size of the matrix I can pass into the SVD? I'm getting seg faults when I try a 100x200 matrix.

[willcannings]

There shouldn't be (other than memory limits). I've used it on larger matrices before without an issue, what OS and architecture are you running on?

[ryana]

It dawned on me that the matrix was under determined: https://gist.github.com/1244393

I'm not a math guy, but I'm guessing that's the problem?

[dchentech]

Here is the same segfault with a small data

require 'rubygems'
require 'ruby-svd'

tdm = SVDMatrix.new(4, 6)
tdm.set_row(0, [5, 5, 3, 0, 5, 5])
tdm.set_row(1, [5, 0, 4, 0, 4, 4])
tdm.set_row(2, [0, 3, 0, 5, 4, 5])
tdm.set_row(3, [5, 4, 3, 3, 5, 5])

puts "== Term document matrix:"
p tdm

puts "\n== Decomposing matrix:"
lsa = LSA.new(tdm)
p lsa

puts "\n== Classifying new column vector: [1, 0.5, 0, 0.5]"
puts "Format is [column, similarity]"
ranks = lsa.classify_vector([1,0.5,0,0.5])
p ranks

sorted_ranks = ranks.sort_by(&:last).reverse
puts "\n== Vector most similar to column #{sorted_ranks.first[0]}"
p tdm.column(sorted_ranks.first[0])

== Term document matrix:
[5, 5, 3, 0, 5, 5]
[5, 0, 4, 0, 4, 4]
[0, 3, 0, 5, 4, 5]
[5, 4, 3, 3, 5, 5]

== Decomposing matrix:
/usr/local/rvm/gems/ree-1.8.7-2011.03/gems/ruby-svd-0.5.1/lib/svd_matrix.rb:41: [BUG] Segmentation fault
ruby 1.8.7 (2011-02-18 patchlevel 334) [i686-darwin10.8.0], MBARI 0x6770, Ruby Enterprise Edition 2011.03

zsh: abort ruby test.rb

[tyrauber]

Any word on the above segmentation fault issue? I would love to use this gem, but I am experiencing the same problem with ruby 1.9.2 on OS X Lion. Seems to crash on anything over 4 columns. Is this an OS X specific issue? Any help would be greatly appreciated.

[CoralineAda]

+1. This is killing me.

[CoralineAda]

After lots of experimentation, I've determined that the segfault occurs when SVDMatrix.new(x, y) where x < y. Any C hackers want to take a crack at this? The problem is a pointer error on line 83 of ext/svd.h

[maxchisto]

I just solved this
https://github.com/maxchisto/Ruby-SVD