deployment - Options for deploying R models in production -


there doesn't seem many options deploying predictive models in production surprising given explosion in big data.

i understand open-source pmml can used export models xml specification. can used in-database scoring/prediction. seems make work need use pmml plugin zementis means solution not open source. there easier open way map pmml sql scoring?

another option use json instead of xml output model predictions. in case, r model sit? i'm assuming need mapped sql...unless r model sit on same server data , run against incoming data using r script?

any other options out there?

the answer depends on production environment is.

if "big data" on hadoop, can try relatively new open source pmml "scoring engine" called pattern.

otherwise have no choice (short of writing custom model-specific code) run r on server. use save save fitted models in .rdata files , load , run corresponding predict on server. (that bound slow can try , throw more hardware @ it.)

how depends on platform. there way add "custom" functions written in r. term udf (user-defined function). in hadoop can add such functions pig (e.g. https://github.com/cd-wood/pigaddons) or can use rhadoop write simple map-reduce code load model , call predict in r. if data in hive, can use hive transform call external r script.

there vendor-specific ways add functions written in r various sql databases. again udf in documentation. instance, postgresql has pl/r.


Comments

Post a Comment

Popular posts from this blog

php - SPIP: From Tag directly to an article -

i work spip , plugin called tagsphere works fine tags in spip. have tag groupe put 1 article tag. when click on tag in tagsphere send me before menu can see articles linked tag. want send me directly article without see tag menu before. here code model of thagsphere: <div id="tagsphere-#env{id_article}"> <ul> <boucle_mot(mots){id_groupe ?}> <li>[<a href="#url_mot">(#titre)</a>]</li> </boucle_mot> </ul> </div> anybody idea? if want skip tag page, have link directly article. <b_article> block prevents display tag if no article linked. <div id="tagsphere-#env{id_article}"> <ul> <boucle_mot(mots){id_groupe ?}> <b_article> <li> <a<boucle_article(articles){...
Read more

jquery - isAjaxRequest always return false -

i got error message 404,'model has not been saved' following code my action controller public function actionextend($id) { $model=$this->loadmodel($id); if(yii::app()->request->isajaxrequest) { $model->todate = date('y-m-d h:i:s a',(strtotime('next month',strtotime(date('y-01-d'))))); if($model->save()) { yii::app()->end(); } else { throw new chttpexception(404,'model has not been saved'); } } } the jquery call $("#buttonextend<?php echo $this->post_row;?>").click(function(e){ $.ajax({ type: "post", url: "<?php echo yii::app()->createurl('post/extend', array('id' => $data['id'])); ?>", }); }); why not ajax call ? thank help i found response, missed } , function ...
Read more

ruby on rails - In a controller spec, how to find a specific tag in the generated view? -

i know integration tests preferred need ran in controller test, i'm testing gem injecting html code in view, xhr can't run in feature spec (if can, please explain me how :) ) so rspec controller tests can assert selector present (with capybara) : response.body.should have_selector('#foobar') has_selector? call all method capybara find selector. what want last child of body , assert id in particular. afaik it's not possible have_selector. what : all('body:first-child').first.id.should == '#foobar' however, capybara dsl, defined (more or less): def all(*args) page.all(*args) end and page empty unless use visit it's integrations specs. how can use capybara all method inside rspec controller test ? i can't test right after googling seems trick def page capybara::node::simple.new(response.body) end source
Read more