Basically it means a model trained with H2O can be transformed to Ignite’s
one and executed in our cluster? Sorry if I’m still missing something.

Denis

On Thursday, October 31, 2019, Alexey Zinoviev <[email protected]>
wrote:

> Thanks, Denis, no, the initial H2O integration is not the same as
> TensorFlow integration, it perform the prediction phase (than you have
> trained model) nor training as I mentioned "As a result we could predict
> something on Ignite data via the model trained in H2O system."
>
> It's the same as XGBost and Spark ML integration, but I hope that we could
> extend our integration here and provide something like Ignite data-source
> for training in H2O in the future.
>
> чт, 31 окт. 2019 г. в 16:48, Denis Magda <[email protected]>:
>
> > Hi Alexey,
> >
> > That’s great to see that our ML and DL capabilities keep growing via
> > integrations!
> >
> > Is it correct to say that H2O uses Ignite as a data source in a way
> similar
> > to what we did for TensorFlow?
> >
> > Btw, do you have any plans to spread the word about the additions through
> > blogging? H2O, Spark ML, XGBoost deserve their own articles.
> >
> > Denis
> >
> > On Thursday, October 31, 2019, Alexey Zinoviev <[email protected]>
> > wrote:
> >
> > > Hi, Igniters,
> > >    I happy to announce the new part of Ignite ML functionality.
> > >
> > > The integration with the popular distributed library H2O
> > > https://github.com/h2oai was added.
> > > As a result we could predict something on Ignite data via the model
> > trained
> > > in H2O system.
> > >
> > > Many thanks to Michal Kurka (https://github.com/michalkurka), waiting
> > him
> > > on our dev-list, hope that he will help with the backward integration
> > from
> > > Ignite to H2O system.
> > >
> > > Remind, that now we have integrations with XGBoost, TensorFlow, Spark
> ML
> > > via SparkMlParser and MLeap integration and waiting for the new
> > > integrations with Dl4j or PyTorch for example
> > >
> > > Sincerely yours,
> > >              Alexey Zinoviev
> > >
> >
> >
> > --
> > -
> > Denis
> >
>


-- 
-
Denis

Reply via email to