Re: Sessionize function - create table or view by default?

2016-07-20 Thread Jim Nasby
On 7/20/16 4:17 PM, Joe Hellerstein wrote: I heartily concur with this assessment FWIW. I generally use views or WITH clauses myself when building up these kinds of pipelines. Note that WITH forces materialization. It's almost as bad as temp tables. In my experience, the PostgreSQL

Re: Sessionize function - create table or view by default?

2016-07-20 Thread Joe Hellerstein
I heartily concur with this assessment FWIW.  I generally use views or WITH clauses myself when building up these kinds of pipelines.  In my experience, the PostgreSQL optimizer is not great at collapsing views though.  So caveat emptor.  Users should be made very aware of EXPLAIN! J On

Re: Sessionize function - create table or view by default?

2016-07-20 Thread Frank McQuillan
Hey Jim, Thank you for the thoughtful response. Given your comments, I think we ought to stick with a view as the default for sessionize. Looking ahead to MADlib 2.0, one thing we want to better support is workflows since chaining operations together is such a common data science thing to do.

Re: Sessionize function - create table or view by default?

2016-07-20 Thread Jim Nasby
On 7/19/16 7:36 PM, Frank McQuillan wrote: "create_view (optional) BOOLEAN default: TRUE. Determines whether to create a view or materialize a table as output. If you only needed session info once, creating a view could be significantly faster than materializing as a table." Question is:

Sessionize function - create table or view by default?

2016-07-19 Thread Frank McQuillan
Hi, I have been testing the sessionize function lately https://issues.apache.org/jira/browse/MADLIB-909 https://issues.apache.org/jira/browse/MADLIB-1001 and am wondering about this param: "create_view (optional) BOOLEAN default: TRUE. Determines whether to create a view or materialize a table