Trying out glmnet: a case study in open-source development

DavidMSmith
Posted by March 12, 2009
Over at Stat Man’s Corner I found a story that really encapsulates one of R’s greatest benefits: the ability to try out new statistical methods and run them through their paces.  Back in January, Trevor Hastie (a well-renowned author and Professor of Statistics at Stanford) announced a new version of glmnet, a machine-learning method for “fitting the entire lasso or elastic-net regularization path for linear regression, logistic and multinomial regression models”.  The details of the model are a bit beyond my ken (see the story for the details of how it was applied to the 1998 Current Population Survey data) — what I find most interesting was the process of getting new types of models in use through R.  The author says it best:

So the R user community had just been provided access to a latest learning algorithm hot off the development presses from three world-renowned practitioners – for free. And glmnet is readily accessible from the internet, installing on existing R platforms painlessly. No commercial stats package that I know of – certainly not the market leader – is even close to releasing a competitive offering. I’d say that’s a pretty good deal for stats types like me, and …

Over at Stat Man’s Corner I found a story that really encapsulates one of R’s greatest benefits: the ability to try out new statistical methods and run them through their paces.  Back in January, Trevor Hastie (a well-renowned author and Professor of Statistics at Stanford) announced a new version of glmnet, a machine-learning method for “fitting the entire lasso or elastic-net regularization path for linear regression, logistic and multinomial regression models”.  The details of the model are a bit beyond my ken (see the story for the details of how it was applied to the 1998 Current Population Survey data) — what I find most interesting was the process of getting new types of models in use through R.  The author says it best:

So the R user community had just been provided access to a latest learning algorithm hot off the development presses from three world-renowned practitioners – for free. And glmnet is readily accessible from the internet, installing on existing R platforms painlessly. No commercial stats package that I know of – certainly not the market leader – is even close to releasing a competitive offering. I’d say that’s a pretty good deal for stats types like me, and a benefit to working with a fertile, world-wide open source initiative like R.

One other point to add: this is also a good example of how early release of new models like this can contribute to their development.  The glmnet package gas been around since at least June 2008 (at least, that’s the date of the oldest version I can find in the CRAN archives). Older versions had some minor problems, but thanks to users trying it out and reporting problems directly back to the author an improved version is available in about 6 months. In his announcement of version 1.1-3, Hastie says:

Thanks to many users, esp. Tim Hesterberg, for notifying us of the errors.

That’s not mere courtesy. That’s a vindication of the open-source process.

Stat Man’s Corner: The R Learning Lasso