Universal Recommender and PredictionIO 0.12.0 incompatibility

[Pat Ferrel]

It looks like PIO 0.12.0 will require a code change in the UR. PIO changed ES1 support drastically when ES5 support was added and broke the UR code.

We will do a quick fix to the template to address this. In the meantime stay on PIO 0.11.0 if you need the UR.

[Vaghawan Ojha]

Thank you for letting us know. 

Thanks

[Noelia Osés Fernández]

Thanks for the heads up!!

[Mars Hall]

Hi Pat,

On 4 October 2017 at 22:04, Pat Ferrel <p...@actionml.com> wrote:

It looks like PIO 0.12.0 will require a code change in the UR. PIO changed ES1 support drastically when ES5 support was added and broke the UR code.

We will do a quick fix to the template to address this. In the meantime stay on PIO 0.11.0 if you need the UR.

We dealt with these breaking Elasticsearch 5.x client changes in our fork of the UR. I've tried to merge these changes back with the main UR source tree, but the ES5 support was not already present so very difficult to pull request. Anyway, we definitely slayed some dragons with the help of Donald Szeto.

What we ended up with is a UR/EsClient that generates its own Elasticsearch RestClient using the Storage config, instead of instantiating PredictionIO's ESStorageClient. This solved a tangled mess of Apache HTTP dependency version conflicts. If you'd like to see what is working well for us (in production, under load), check out this [merged] PR to our fork:

https://github.com/heroku/predictionio-engine-ur/pull/7/files

It would be wonderful to have your main UR working with this ES5 capability. Let me know if you have questions about our approach,

--

*Mars Hall

415-818-7039

Customer Facing Architect

Salesforce Platform / Heroku

San Francisco, California

[Pat Ferrel]

Yes, easy enough to do but rather annoying. Rather than supporting 1.x and 5.x I think the UR 0.6.1 will be EOL for ES 1.x and UR 0.7.0 will be ES 5.x from then on. We have a major speedup in 0.7.0 and an even greater speedup in the next after that, which will start using Spark dataframes instead of Mahout Matrices and BiMaps. We are also starting to design a Kappa style version of the UR that will be based on Harness instead of PredictionIO. It won’t require a training phase or Spark at all if you don’t need to bootstrap from old data.

--
You received this message because you are subscribed to the Google Groups "actionml-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to actionml-use...@googlegroups.com.
To post to this group, send email to action...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CA%2BVXBtsA_C%2B_jqLw85tOZQbFwsdVM4Rb1Xhu_Fp_8%2BhX3ukEow%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

[bernhard...@maxodus.net]

Hi Pat, 

Am Mittwoch, 4. Oktober 2017 22:04:32 UTC+2 schrieb Pat Ferrel:

It looks like PIO 0.12.0 will require a code change in the UR. PIO changed ES1 support drastically when ES5 support was added and broke the UR code.

We will do a quick fix to the template to address this. In the meantime stay on PIO 0.11.0 if you need the UR.

Do you have any ETA (days, weeks or months) on that quick fix? I got PIO 0.12.0 running nicely and would prefer not to downgrade for UR.

Cheers, 

Bernhard 

[Pat Ferrel]

Working on it now. I may get something pushed today or tomorrow. Would love to have a tester. I’ll post here when its ready to look at.

 

--
You received this message because you are subscribed to the Google Groups "actionml-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to actionml-use...@googlegroups.com.
To post to this group, send email to action...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/255be6a4-033e-4289-b53c-605430f8e5c6%40googlegroups.com.

[Bernhard Hortens]

That sounds great!

Shoot me a message, I’ll be happy to test!

Cheers,

Bernhard

You received this message because you are subscribed to a topic in the Google Groups "actionml-user" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/actionml-user/VgoFe_qf9a4/unsubscribe.
To unsubscribe from this group and all its topics, send an email to actionml-use...@googlegroups.com.

To post to this group, send email to action...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/2A979099-973C-453F-97BC-57FA5592C0BE%40occamsmachete.com.