Ignite ML dense distributed matrices

[ArtemM]

Hi all, currently we do not have dense distributed matices in ML module, but
I think we should add them. Obviously, not every big  ML dataset has sparse
structure. Of course, we can put it into a sparse distributed matrix, but
this will have performance hits. Any ideas/objections about it?



--
Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/

[Yuriy Babak]

Artem,

I think It`s a good idea. We could implement dense matrix as separate
matrix, but what do you think about common distributed matrix with multiple
possible storage strategies?

Regards,
Yury



--
Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/

[Alexey Zinoviev]

Yury, I support one matrix with different strategies if it will have one
API.
Also distributed/local can be a strategy too.

And yet, Artem, Yury, we could think about DenceBlockDistributedMatrix of
course.

Or about Matrix(boolean isDistributed, boolean isBlock, boolean isSparse)
and implement all 8 cases

2017-11-20 18:24 GMT+03:00 Yury Babak <[hidden email]>:

> Artem,
>
> I think It`s a good idea. We could implement dense matrix as separate
> matrix, but what do you think about common distributed matrix with multiple
> possible storage strategies?
>
> Regards,
> Yury
>
>
>
> --
> Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/
>

[Yuriy Babak]

Alexey,

I dont think that mixing distributed/local is a good idea unlike
dense/sparse or blocks/rows/cols.

Regards,
Yury



--
Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/