[GitHub] spark issue #23146: [SPARK-26173] [MLlib] Prior regularization for Logistic ...

[elfausto <gi...@git.apache.org>]

Github user elfausto commented on the issue:

    https://github.com/apache/spark/pull/23146
  
    Thanks for your review, @srowen.
    
    I added a link to the PR's description and the JIRA ticket pointing to the reference Python implementation I used for unit testing.
    
    Prior regularization fits nicely with taking a Bayesian approach to Logistic Regression by posing a Gaussian prior and taking a Laplace approximation to the posterior. I'm not sure about approaching other regularized models in the same fashion.
    
    As a reference, besides this being a textbook algorithm, the original motivation for its implementation at [Affectv](https://affectv.com/), was the following paper showing its application to display advertising: _Olivier Chapelle , Eren Manavoglu , Romer Rosales, Simple and Scalable Response Prediction for Display Advertising, ACM Transactions on Intelligent Systems and Technology (TIST), v.5 n.4, January 2015_.


---

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