Can classifier-reborn work with Numo::NArray / Numo::GSL ? Is that a better choice than nmatrix?

[0xdevalias]

I noticed that https://jekyll.github.io/classifier-reborn/#dependencies states we can use the rb-gsl gem to speed things up, and that in turn references using nmatrix or narray.

The linked narray GitHub page states that it is in maintenance mode, and links to Numo::NArray, which lists Numo::GSL under related projects.

Is it possible to use Numo::GSL with classifier-reborn?

---

Since rb-gsl can only use one of nmatrix or narray, i'm also wondering which would be the better (aka: faster) choice to use for this functionality?

In order to use rb-gsl with NMatrix you must first set the NMATRIX environment variable and then install rb-gsl: gem install nmatrix export NMATRIX=1 gem install rb-gsl

This will compile rb-gsl with NMatrix specific functions.

For using rb-gsl with NArray: gem install narray export NARRAY=1 gem install rb-gsl

Note that setting both NMATRIX and NARRAY variables will lead to undefined behaviour. Only one can be used at a time.

I also noticed that rb-gsl hasn't been updated since 2017, and so is most likely unmaintained:

• Is this project still maintained? SciRuby/rb-gsl#63

[mkasberg]

Bringing this issue back to life! I love Jekyll ❤️ and still use it for several personal projects. I also use this classifier-reborn gem, and it would be great to make sure it remains easy to use with more modern hardware & software. Here are some notes about why this matters:

• This gem is currently testing in CI against Ruby 2.4 and 2.6, both of which are EOL.

  classifier-reborn/.travis.yml

  Lines 4 to 5 in c0e81d2

  	- 2.4
  	- 2.6

• Ruby 2.7 will be the last 2.x branch, and will be EOL in less than a year.
• Ubuntu 22.04 no longer includes openssl 1.1, making it difficult to build/use 2.x branches of Ruby.
• Likewise, my understanding is that using Ruby with openssl 1.1 via brew on macOS is possible, but is not the default and requires extra configuration to build.

All that is to say... it's increasingly important that this gem is easy to use with Ruby 3. However:

• Using classifier-reborn on my ~50 post blog is unusable (hours of build time) without GSL.
• rb-gsl is unmaintained - luckily, there's a commit on main that makes it compatible with Ruby 3, but nobody has released the gem so the only way to use rb-gsl with Ruby 3 right now is to specify the git hash in your Gemfile. See Publish a new release to rubygems SciRuby/rb-gsl#67
• As pointed out above, rb-gsl is itself using a deprecated/unmaintained narray implementation.

I would like to write some PRs to try to:

• Update CI to work with Ruby 2.7 and 3.x on GitHub Actions so we have good CI/tests with modern Ruby versions.
• Add Numo::Linalg as an alternative to rb-gsl. This could work similarly to the way $GSL currently works, detecting the availability at runtime.
• Ultimately, produce a released gem version that's compatible with Ruby 3.x in a way that does not require using unreleased versions of dependencies for decent performance.

I've experimented a little bit (essentially a prototype/spike) and I think I can get a working SVD using Numo::Linalg. For example, here's an alternative implementation of LSI#build_reduced_matrix I was experimenting with:

    def numo_build_reduced_matrix(matrix, cutoff = 0.75)
      # OPTIMIZE ME: Consider other drivers/options like sdd.
      s, u, vt = Numo::Linalg.svd(matrix, driver: 'svd', job: 'S')

      # TODO: Better than 75% term, please. :\
      s_cutoff = s.sort.reverse[(s.size * cutoff).round - 1]
      s.size.times do |ord
        s[ord] = 0.0 if s[ord] < s_cutoff
      end

      # Reconstruct the term document matrix, only with reduced rank
      u.dot(::Numo::DFloat.eye(s.size) * s).dot(vt)
    end

To make this work and get it merged, I need the support of a maintainer on this project. Is there anyone in here who could review & merge PRs for the above? I realize this is an open source side project for most people (me too!) and I'm not looking for immediate response/turnaround. At the same time, I'm hesitant to invest in this work if my PRs will sit for months with no response.

[mattr-]

If you can make the PRs, I can review and merge them.

[mkasberg]

@mattr- what's the process like for releasing a new gem version now that #198 is merged?

[mattr-]

@mkasberg will research and find out. It's likely similar to what we have in the maintainer instructions for the core jekyll gem.

[mattr-]

I've pushed a v2.3.0 tag and am waiting on permission changes to rubygems.org to push the new gem itself. There will be a new release soon™

[mattr-]

Since v2.3.0 will support Numo, I'm going to go ahead and close this.

[mkasberg]

@mattr- Any updates? I still don't see 2.3.0 at https://rubygems.org/gems/classifier-reborn