Development

[dselivanov]

@cdeterman I can try to give a shot on this. I just may need help with initial set-up of the repository - how to link to vienacl, what is the preferred structure of S4 classes, etc.

[cdeterman]

@dselivanov thanks for the offer. I can put together a general skeleton that we can start to work from.

[cdeterman]

@dselivanov A very minimal start at the moment. I wanted to push it up as I have encountered a silly error when using devtools::test() where my S4 typeof method is failing. I am also tracking down a ViennaCL related issue with the shared_ptr useage. It is at least a start.

[dselivanov]

Thanks you @cdeterman ! Now I believe at least we can work asynchronously. Not sure when I actually will be able to start with this, but it is something which I want to give a shot.

[cdeterman]

@dselivanov I am actually stuck at the moment dealing with that S4 issue. I have posted an SO question that hopefully some kind soul can lend some insight. It is a curious issue with S4 methods in multiple packages.

[dselivanov]

Mb it make sense to add sparse matrices to main gpuR repo? чт, 21 июн. 2018 г., 17:59 Charles Determan <[email protected]>:

…

@dselivanov <https://github.com/dselivanov> I am actually stuck at the moment dealing with that S4 issue. I have posted an SO question <https://stackoverflow.com/questions/50952511/s4-method-fails-when-also-set-in-another-package> that hopefully some kind soul can lend some insight. It is a curious issue with S4 methods in multiple packages. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AE4u3aeNC27Vl3XAp0LS0ITQUe8dMVX8ks5t-6bPgaJpZM4UfoUv> .

[cdeterman]

@dselivanov the problem was coming because typeof is not a generic function. I am now creating a separate generics package that creates general purpose generics to resolve this issue across packages. Once everything is stable there, I will submit that to CRAN as well. I have confirmed that this solves the problem locally.

Regarding your previous point, I don't want to bloat the gpuR package. It is already a pretty large package and sparse operations may be more of a specialty thing. I would like to maintain that separately.

[jeffkeller87]

@cdeterman just wondering if this is still being pursued. The gpuR package is just amazing and I'd love to be able to use it for sparse matrices as well.

[cdeterman]

@jeffkeller87 I certainly want to, I simply don't have the bandwidth currently. I would love to have a collaborator to get this moving along further. But currently it is stalled.

[jeffkeller87]

Thanks for the quick reply @cdeterman. I am interested in collaborating--what's the best way to get in touch? Is that here in GitHub Issues? Signal? Riot/Matrix? I'd like to better understand the current state of gpuRsparse and what tasks you see to get it to a state worthy of a CRAN submission.

[cdeterman]

@jeffkeller87 sorry for delayed response, I have been quite busy lately.

I have set up a gitter.im room here where we can have general discussions. Naturally, issues should be created for anything concrete.

The general state of gpuRsparse is essentially non-functional. I simply have the outline for it formed but there should be much to be leveraged directly from ViennaCL. The main issue I initially ran in to was dealing with the S4 generic problem. I have seen that there is a generics package with CRAN so I will need to change my proposed package name before I can submit that.

Afterwards, it will essentially be working on implementing all the basic operations (e.g. addition, subtraction, etc.) in addition to working on the sparse solvers that are part of ViennaCL.

I am still struggling to get gpuR accepted once again currently. For some reason the build is failing on CRAN that I cannot replicate. Hopefully it is resolved soon and I can find some more time to dedicate to these packages.