Thanks Rong, really appreciate review.
See some answers below
Boris Lublinsky
FDP Architect
boris.lublin...@lightbend.com
https://www.lightbend.com/
> On May 5, 2019, at 7:52 PM, Rong Rong wrote:
>
> Hi Robert, Boris,
>
> Sorry for the late reply. I took some time to look at the implementatio
Hi Robert, Boris,
Sorry for the late reply. I took some time to look at the implementation. I
think it looks good overall to me.
Since this is pretty much a big PR. I would like to raise a bit of
discussion beforehand:
1. In order to simplify the PR complexity, can we do the following?
- the exam
Hey all,
I'm wondering if somebody on the list can take a look at the PR from
FLIP-23: https://github.com/apache/flink/pull/7446
On Mon, Oct 1, 2018 at 6:13 PM Rong Rong wrote:
> Thanks for the contribution Boris!! I've been playing around with the basic
> model for a while back and loved it.
Thanks for the contribution Boris!! I've been playing around with the basic
model for a while back and loved it.
+1 and really looking forward to having the feature merging back to Flink
ML.
--
Rong
On Mon, Oct 1, 2018 at 7:55 AM Fabian Hueske wrote:
> Hi everybody,
>
> The question of how to s
Hi everybody,
The question of how to serve ML models in Flink applications came up in
several conversations I had with Flink users in the last months.
Recently, Boris approached me and he told me that he'd like to revive the
efforts around FLIP-23 [1].
In the last days, Boris extended the proposa
Thanx @Fabian. I will update the document accordingly wrt metrics.
I agree there are pros and cons.
Best,
Stavros
On Wed, Jan 31, 2018 at 1:07 AM, Fabian Hueske wrote:
> OK, I think there was plenty of time to comment on this FLIP.
> I'll move it to the ACCEPTED status.
>
> @Stavros, please co
OK, I think there was plenty of time to comment on this FLIP.
I'll move it to the ACCEPTED status.
@Stavros, please consider the feedback regarding the metrics.
I agree with Chesnay that metrics should be primarily exposed via the
metrics system.
Storing them in state makes them fault-tolerant and
I'm currently looking over it, but one thing that stood out was that the
FLIP proposes to use queryable state
as a monitoring solution. Given that we have a metric system that
integrates with plenty of commonly used
metric backends this doesn't really make sense to me.
Storing them in state sti
Are there any more comments on the FLIP?
Otherwise, I'd suggest to move the FLIP to the accepted FLIPs [1] and
continue with the implementation.
Also, is there a committer who'd like to shepherd the FLIP and review the
corresponding PRs?
Of course, everybody is welcome to review the code but we n
Hi,
Sorry for the late follow up.
I think I understand the motivation for choosing ProtoBuf as the
representation and serialization format and this makes sense to me.
However, it might be a good idea to provide tooling to convert Flink types
(described as TypeInformation) to ProtoBuf.
Otherwise,
Hi Boris and Stavros,
Thanks for the responses.
Ad 1) Thanks for the clarification. I think I misunderstood this part of
the proposal.
I interpreted the argument why to chose ProtoBuf for network encoding ("ability
to represent different data types") such that different a model pipeline
should wo
Hi Fabian thanx!
> 1) Is it a strict requirement that a ML pipeline must be able to handle
> different input types?
> I understand that it makes sense to have different models for different
> instances of the same type, i.e., same data type but different keys. Hence,
> the key-based joins make se
Hi Stavros,
thanks for the detailed FLIP!
Model serving is an important use case and it's great to see efforts to add
a library for this to Flink!
I've read the FLIP and would like to ask a few questions and make some
suggestions.
1) Is it a strict requirement that a ML pipeline must be able to
Hi guys,
Let's discuss the new FLIP proposal for model serving over Flink. The idea
is to combine previous efforts there and provide a library on top of Flink
for serving models.
https://cwiki.apache.org/confluence/display/FLINK/FLIP-23+-+Model+Serving
Code from previous efforts can be found her
14 matches
Mail list logo
