[
https://issues.apache.org/jira/browse/SPARK-11988?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15030469#comment-15030469
]
Villu Ruusmann commented on SPARK-11988:
----------------------------------------
I went through JPMML-Model commit log and here's my take on "significant
functional changes" between 1.1.15 and 1.2.7:
* 1.2.0. Renamed Fluent API-style setter methods to POJO-style setter methods.
For example, PMML#withVersion(String) became PMML#setVersion(String).
* 1.2.1. Included modified PMML XSD resource (adds removed and deprecated 3.X
and 4.X elements and attributes to the latest 4.2.1 XSD resource) to
distributable JAR file. You can use standard JAXB APIs to validate newly
generated PMML documents against this XSD. For example, this would have caught
the missing PMML@version attribute easily.
* 1.2.2. Activated the emulation of JAXB "lax" wildcard syntax. It is about
parsing/formating an PMML document "deeply", when it contains mixed PMML and
custom markup.
* 1.2.3 and 1.2.4. Reorganized org.dmg.pmml.Field class hierarchy. There is a
distinction between "field definition" and "field reference" elements now. For
example, DataField is a field definition element, whereas MiningField is merely
a field reference element. Added org.dmg.pmml.Indexable interface, which marks
unique element/attribute combinations. For example, it provides an easy way to
force that every DataField element in a DataDictionary element has a unique
DataField@name attribute.
* 1.2.5. Generated "test-if-exists-and-is-not-empty" test methods for
Collection-type class model attributes. The idea is to avoid accessing
missing/empty Collection attributes via getter methods, because that would
allocate an unnecessary ArrayList instance. For example, one should test
Node#hasNodes() before accessing Node#getNodes() for real work.
* 1.2.6. Fixed a concurrency bug about FieldName interning.
* 1.2.7. Added a bunch of HasExpression, HasPredicate, HasValue, HasValueSet
etc. marker interfaces that come in handy when rearranging PMML class model
objects using Visitor API.
> Update JPMML to 1.2.7
> ---------------------
>
> Key: SPARK-11988
> URL: https://issues.apache.org/jira/browse/SPARK-11988
> Project: Spark
> Issue Type: Improvement
> Components: ML, MLlib
> Affects Versions: 1.5.2
> Reporter: Sean Owen
> Assignee: Sean Owen
> Priority: Minor
>
> Pretty much says it all: update to the more recent JPMML 1.2.x line from
> 1.1.x.
> The API did change between the 2 versions, though all use of it is internal
> to Spark, so doesn't affect end users in theory.
> In practice, JPMML leaks into the classpath and might cause issues. My
> rationale for the change is that 1.2.x is more recent and more stable (right
> [~vfed]?) so if people are going to run into it, might as well run into
> something more modern.
> And bug fixes and more support and all that.
> CC [~selvinsource]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
