JOML rotate ops

[rimadoma]

• Adds LinAlg namespace
• Add JOML dependency
• Adds rotation ops for Vector3d and Vector3f
• Adds various convenience OpMethods for the above
• Bumps SNAPSHOT version

[rimadoma]

If this PR is accepted, maybe the linalg namespace would be a good place for the op in PR #420

[ctrueden]

Cool, thanks @rimadoma! Do you think it would be too general to call the namespace linear instead of linalg? What do other software libraries call it?

[rimadoma]

I was thinking this name space would be for linear algebra on (simple) vectors and matrices. I think linear could easily lead to a loss of coherence, where the name space hosts ops from vastly different domains of maths and image processing. How about just calling it vecmath?

Apache commons calls their package org.apache.commons.math3.linear, but I think their situation is different, since the package is just limited to mathematics.

In any case, I see these ops eventually migrating to scijava-ops at which point the naming can be reconsidered.

[imagejan]

linalg is used by numpy/scipy as well, and in MATLAB (at least in their outdated MuPAD notebooks), it seems.

I think linalg isn't such a bad choice.

[bnorthan]

What do other software libraries call it?

@ctrueden in the last year or so I've mostly been wrapping BLAS (Basic Linear Algebra Specification), and using a bit of ND4J (N-Dimensionsal Arrays for Java). I'd vote for calling the namespace linalg and use it to implement a subset of blas functionality in a user friendly way.

@rimadoma if you haven't seen it already there is a transform namespace that contains a rotate function. Although this namespace is specific to imglib2 structures.

I've also been working a RealTransform to apply a affine transform to an image.

Do you guys think all rotate ops should be in transform or is it OK to have a rotate in linalg ??

I also wonder if this functionality belongs in the core imagej-ops project, or should be in a separate joml-ops repo. Should imagej-ops be for imglib2 implementations, and other implementations implemented in separate projects??

[xulman]

Do you guys think all rotate ops should be in transform or is it OK to have a rotate in linalg ??

Do we have/Can we have aliases in the ops world?
It is not always black-and-white to categorize stuff...

I also prefer linalg.

[rimadoma]

@bnorthan All the ops in the transform namespace seem to handle RAIs, so imo adding Vector3d stuff would make it less coherent. Maybe it is worth then, like you suggested, to have a joml-ops or other such separate module, and limit imagej-ops to imglib2 things.

[ctrueden]

This imagej/imagej-ops repository needs to be chopped into multiple different repositories before year's end. I agree that putting the JOML ops into their own repository makes sense. Calling the namespace linalg sounds good, since it sounds more consistent and preferred by all. Regarding aliases, yes, ops can have any number of aliases, including aliases in different namespaces. But we should choose them carefully because we will need to support them all into the future—if we want to reclaim a particular alias later for something different, it would still break existing scripts expecting that op to be something else.

@rimadoma I am going on vacation starting Wednesday for two weeks, so will probably not get to this PR before then. But I will try to make it a priority to merge it before end of August.

[rimadoma]

What's your opinion y'all: should this linalg repository also host ojAlgo i.e. arbitrary size matrix ops as well? The consensus at least at the MPI-CBG Dresden '17 hackathon was that there's need for both JOML and ojAlgo libraries for linear algebra.

[rimadoma]

@ctrueden Enjoy your vacation! I'll start mine before you come back, and won't return until August 2.

[ctrueden]

This is a beautiful PR, so I'm going to merge it. We can deal with the modularization separately as part of the more general modularization of ImageJ Ops.