[jira] [Comment Edited] (SPARK-18213) Syntactic sugar over Pipeline API

["Wojciech Szymanski (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-18213?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15635662#comment-15635662 ] 

Wojciech Szymanski edited comment on SPARK-18213 at 11/4/16 8:33 AM:
---------------------------------------------------------------------

Thanks for your opinion. Initially I was thinking about varargs based constructor, since stage array is the only one attribute supported by pipeline. 
{code}
// only Scala
val pipeline = new Pipeline(tokenizer, stopWordsRemover, countVectorizer)
{code}
Unfortunately, current Scala compiler does not support generating pure Java varargs constructors with @varargs annotation. 

Another option is companion object, but again, it wouldn't be convenient from Java perspective.
{code}
// Scala
val pipeline = Pipeline(tokenizer, stopWordsRemover, countVectorizer)
// Java  - ugly approach
Pipeline pipeline = Pipeline.apply(tokenizer, stopWordsRemover, countVectorizer);
{code}

Last thing that comes to my mind is array based constructor, but on the other hand it does not simplify much.
{code}
// Scala
val pipeline = new Pipeline(Array(tokenizer, stopWordsRemover, countVectorizer))
// Java 
Pipeline pipeline = new Pipeline(new Pipeline[] {tokenizer, stopWordsRemover, countVectorizer});
{code}
{code}


was (Author: wojtek-szymanski):
Thanks for your opinion. Initially I was thinking about varargs based constructor, since stage array is the only one attribute supported by pipeline. 
{code}
// only Scala
val pipeline = new Pipeline(tokenizer, stopWordsRemover, countVectorizer)
{code}
Unfortunately, current Scala compiler does not support generating pure Java varargs constructors with @varargs annotation. 

Another option is companion object, but again, it wouldn't be convenient from Java perspective.
{code}
// Scala
val pipeline = Pipeline(tokenizer, stopWordsRemover, countVectorizer)
// Java  - ugly approach
Pipeline pipeline = Pipeline.apply(tokenizer, stopWordsRemover, countVectorizer);
{code}

Last thing that comes to my mind is array based constructor, but on the other hand it does not simplify much.
{code}
// Scala
val pipeline = new Pipeline(Array(tokenizer, stopWordsRemover, countVectorizer))
// Java 
Pipeline pipeline = Pipeline.apply(new Pipeline[] {tokenizer, stopWordsRemover, countVectorizer});
{code}
{code}

> Syntactic sugar over Pipeline API
> ---------------------------------
>
>                 Key: SPARK-18213
>                 URL: https://issues.apache.org/jira/browse/SPARK-18213
>             Project: Spark
>          Issue Type: Improvement
>          Components: ML
>    Affects Versions: 2.0.1
>            Reporter: Wojciech Szymanski
>            Priority: Minor
>
> Currently, creating ML Pipeline is based on very verbose setStages method as below:
> {code}
>     val tokenizer = new RegexTokenizer()
>     val stopWordsRemover = new StopWordsRemover()
>     val countVectorizer = new CountVectorizer()
>     val pipeline = new Pipeline().setStages(Array(tokenizer, stopWordsRemover, countVectorizer))
> {code}
> What about a bit of syntactic sugar over Pipeline API?
> {code}
>     val tokenizer = new RegexTokenizer()
>     val stopWordsRemover = new StopWordsRemover()
>     val countVectorizer = new CountVectorizer()
>     val pipeline = tokenizer + stopWordsRemover + countVectorizer
> {code}
> Production code changes in mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala:
> https://github.com/apache/spark/commit/181df64bf50081f3af5a84b567b677178c88524f#diff-5226e84dea43423760dc6300ddafb01b
> Scala example:
> https://github.com/apache/spark/commit/181df64bf50081f3af5a84b567b677178c88524f#diff-798e85dd9107565fabab1126f57e3d6e
> Java example:
> https://github.com/apache/spark/commit/181df64bf50081f3af5a84b567b677178c88524f#diff-69ac857220f21b5e1684444d80d6dffe
> Thanks in advance for your feedback.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org