Ignite ML Models improvements

[Yuriy Babak]

Hi all.

I want to improve our models. Currently our models can predict result and be
combined with other models, thats all. So I will add import/export
functionality, check serialization performance and add some related tests.

And also I want to ask community - may be we need something else from our
models?

Thanks,
Yury



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

[Vladisav Jelisavcic]

Hi Yury,

since we're already on topic for importing/exporting models,
just as a thought: maybe we should add support for PMML [1] [2]?

Best regards,
Vladisav

[1] http://dmg.org/pmml/v4-1/GeneralStructure.html
[2] https://github.com/jpmml/jpmml-model

On Mon, Oct 16, 2017 at 2:20 PM, Yury Babak <[hidden email]> wrote:

> Hi all.
>
> I want to improve our models. Currently our models can predict result and
> be
> combined with other models, thats all. So I will add import/export
> functionality, check serialization performance and add some related tests.
>
> And also I want to ask community - may be we need something else from our
> models?
>
> Thanks,
> Yury
>
>
>
> --
> Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/
>

[ArtemM]

Hi, Vladislav,

I'm not familiar with PMML, but quick googling has shown that Tensorflow,
Keras, DL4j and some other major contemporary ML frameworks do not have
support of it (at least native). So the question is why should we include
it, can you please elaborate?

Best regards, Artem.



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

[Vladisav Jelisavcic]

Hi Artem,

I agree, it is not supported in many frameworks;
on the other hand, Spark MLlib supports it [1].

I think it is good thing to check with users if there is a need for this,
since we are already on the topic of model importing/exporting.

Best regards,
Vladisav

[1] https://spark.apache.org/docs/2.1.1/mllib-pmml-model-export.html



On Mon, Oct 16, 2017 at 2:57 PM, ArtemM <[hidden email]> wrote:

> Hi, Vladislav,
>
> I'm not familiar with PMML, but quick googling has shown that Tensorflow,
> Keras, DL4j and some other major contemporary ML frameworks do not have
> support of it (at least native). So the question is why should we include
> it, can you please elaborate?
>
> Best regards, Artem.
>
>
>
> --
> Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/
>

[Yuriy Babak]

Vladisav,

Thanks for this idea. I`m not sure that this is priority task but anyway I
created separate ticket for it(IGNITE-6642).

Regards,
Yury.





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