spark-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Joseph Bradley <jos...@databricks.com>
Subject Re: Revisiting Online serving of Spark models?
Date Mon, 21 May 2018 20:52:21 GMT
Regarding model reading and writing, I'll give quick thoughts here:
* Our approach was to use the same format but write JSON instead of
Parquet.  It's easier to parse JSON without Spark, and using the same
format simplifies architecture.  Plus, some people want to check files into
version control, and JSON is nice for that.
* The reader/writer APIs could be extended to take format parameters (just
like DataFrame reader/writers) to handle JSON (and maybe, eventually,
handle Parquet in the online serving setting).

This would be a big project, so proposing a SPIP might be best.  If people
are around at the Spark Summit, that could be a good time to meet up & then
post notes back to the dev list.

On Sun, May 20, 2018 at 8:11 PM, Felix Cheung <felixcheung_m@hotmail.com>
wrote:

> Specifically I’d like bring part of the discussion to Model and
> PipelineModel, and various ModelReader and SharedReadWrite implementations
> that rely on SparkContext. This is a big blocker on reusing  trained models
> outside of Spark for online serving.
>
> What’s the next step? Would folks be interested in getting together to
> discuss/get some feedback?
>
>
> _____________________________
> From: Felix Cheung <felixcheung_m@hotmail.com>
> Sent: Thursday, May 10, 2018 10:10 AM
> Subject: Re: Revisiting Online serving of Spark models?
> To: Holden Karau <holden@pigscanfly.ca>, Joseph Bradley <
> joseph@databricks.com>
> Cc: dev <dev@spark.apache.org>
>
>
>
> Huge +1 on this!
>
> ------------------------------
> *From:* holden.karau@gmail.com <holden.karau@gmail.com> on behalf of
> Holden Karau <holden@pigscanfly.ca>
> *Sent:* Thursday, May 10, 2018 9:39:26 AM
> *To:* Joseph Bradley
> *Cc:* dev
> *Subject:* Re: Revisiting Online serving of Spark models?
>
>
>
> On Thu, May 10, 2018 at 9:25 AM, Joseph Bradley <joseph@databricks.com>
> wrote:
>
>> Thanks for bringing this up Holden!  I'm a strong supporter of this.
>>
>> Awesome! I'm glad other folks think something like this belongs in Spark.
>
>> This was one of the original goals for mllib-local: to have local
>> versions of MLlib models which could be deployed without the big Spark JARs
>> and without a SparkContext or SparkSession.  There are related commercial
>> offerings like this : ) but the overhead of maintaining those offerings is
>> pretty high.  Building good APIs within MLlib to avoid copying logic across
>> libraries will be well worth it.
>>
>> We've talked about this need at Databricks and have also been syncing
>> with the creators of MLeap.  It'd be great to get this functionality into
>> Spark itself.  Some thoughts:
>> * It'd be valuable to have this go beyond adding transform() methods
>> taking a Row to the current Models.  Instead, it would be ideal to have
>> local, lightweight versions of models in mllib-local, outside of the main
>> mllib package (for easier deployment with smaller & fewer dependencies).
>> * Supporting Pipelines is important.  For this, it would be ideal to
>> utilize elements of Spark SQL, particularly Rows and Types, which could be
>> moved into a local sql package.
>> * This architecture may require some awkward APIs currently to have model
>> prediction logic in mllib-local, local model classes in mllib-local, and
>> regular (DataFrame-friendly) model classes in mllib.  We might find it
>> helpful to break some DeveloperApis in Spark 3.0 to facilitate this
>> architecture while making it feasible for 3rd party developers to extend
>> MLlib APIs (especially in Java).
>>
> I agree this could be interesting, and feed into the other discussion
> around when (or if) we should be considering Spark 3.0
> I _think_ we could probably do it with optional traits people could mix in
> to avoid breaking the current APIs but I could be wrong on that point.
>
>> * It could also be worth discussing local DataFrames.  They might not be
>> as important as per-Row transformations, but they would be helpful for
>> batching for higher throughput.
>>
> That could be interesting as well.
>
>>
>> I'll be interested to hear others' thoughts too!
>>
>> Joseph
>>
>> On Wed, May 9, 2018 at 7:18 AM, Holden Karau <holden@pigscanfly.ca>
>> wrote:
>>
>>> Hi y'all,
>>>
>>> With the renewed interest in ML in Apache Spark now seems like a good a
>>> time as any to revisit the online serving situation in Spark ML. DB &
>>> other's have done some excellent working moving a lot of the necessary
>>> tools into a local linear algebra package that doesn't depend on having a
>>> SparkContext.
>>>
>>> There are a few different commercial and non-commercial solutions round
>>> this, but currently our individual transform/predict methods are private so
>>> they either need to copy or re-implement (or put them selves in
>>> org.apache.spark) to access them. How would folks feel about adding a new
>>> trait for ML pipeline stages to expose to do transformation of single
>>> element inputs (or local collections) that could be optionally implemented
>>> by stages which support this? That way we can have less copy and paste code
>>> possibly getting out of sync with our model training.
>>>
>>> I think continuing to have on-line serving grow in different projects is
>>> probably the right path, forward (folks have different needs), but I'd love
>>> to see us make it simpler for other projects to build reliable serving
>>> tools.
>>>
>>> I realize this maybe puts some of the folks in an awkward position with
>>> their own commercial offerings, but hopefully if we make it easier for
>>> everyone the commercial vendors can benefit as well.
>>>
>>> Cheers,
>>>
>>> Holden :)
>>>
>>> --
>>> Twitter: https://twitter.com/holdenkarau
>>>
>>
>>
>>
>> --
>>
>> Joseph Bradley
>>
>> Software Engineer - Machine Learning
>>
>> Databricks, Inc.
>>
>> [image: http://databricks.com] <http://databricks.com/>
>>
>
>
>
> --
> Twitter: https://twitter.com/holdenkarau
>
>
>


-- 

Joseph Bradley

Software Engineer - Machine Learning

Databricks, Inc.

[image: http://databricks.com] <http://databricks.com/>

Mime
View raw message