Remove beta warning lazy signal

[ericpre]

The lazy signal API has been used for several years and it is mature enough: it is stable and as of today, there is no reason that there will significant change any time soon in the API! Therefore, I think that it is fair to remove this warning! @francisco-dlp, @CSSFrancis, what do you think?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (c9f4a53) 80.25% compared to head (e848200) 80.25%.

Additional details and impacted files

@@                 Coverage Diff                 @@
##           RELEASE_next_major    #3282   +/-   ##
===================================================
  Coverage               80.25%   80.25%           
===================================================
  Files                     147      147           
  Lines                   21854    21854           
  Branches                 5139     5139           
===================================================
  Hits                    17538    17538           
  Misses                   3111     3111           
  Partials                 1205     1205           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[CSSFrancis]

@ericpre that sounds like a great idea. One thing to note is that distributed computing isn't really fully supported and should probably remain in beta.

I thought I made an issue tracker for that but I don't know if I got around to it. As far as I know the samfire and matrix factorization (as well as downstream things like orientation mapping) are what isn't currently supported using lazy processing. In addition most file types aren't supported either (but I'm slowly adding more and more support)

[ericpre]

Yes, good point, instead of removing the warning, we should update it to be more specific.

• distributed computing: only a limited number of file format are supported and we should signpost to https://hyperspy.org/rosettasciio/user_guide/supported_formats/index.html
• matrix factorization: interesting, I was expecting that it would work because I thought that this is fairly standard things but I haven't tried and maybe I am being a bit naive!

[francisco-dlp]

I agree that the learning features should remain in beta. They still need a lot of polishing and some basic functionality, such as SVD, work lazily in lazy signals, which sounds right, but that is often not what the user needs. In particular, if things haven't changed since I last used the feature, SVD computes the components on demand.

[CSSFrancis]

Yes, good point, instead of removing the warning, we should update it to be more specific.

• distributed computing: only a limited number of file format are supported and we should signpost to https://hyperspy.org/rosettasciio/user_guide/supported_formats/index.html

That sounds good to me!

• matrix factorization: interesting, I was expecting that it would work because I thought that this is fairly standard things but I haven't tried and maybe I am being a bit naive!

I agree that the learning features should remain in beta. They still need a lot of polishing and some basic functionality, such as SVD, work lazily in lazy signals, which sounds right, but that is often not what the user needs. In particular, if things haven't changed since I last used the feature, SVD computes the components on demand.

I think you are right. I also think that the NNMF can have some problems with large datasets (if I remember). It needs to be rewritten to use map_blocks rather than the _block_iterator function. It actually wouldn't be too hard but I don't really use it so I've never devoted the time to re-writing it. It probably would be a good thing to do as there really aren't lazy nnmf codes in python. See this surprisingly empty issue :) dask/dask-ml#96.

That code should probably make its way into something like dask-ml.

[CSSFrancis]

This looks good to me!