The next meeting will be held on Wednesday 21st of September at 5.30pm in Room LG.11 in David Hume Tower.
Please note a change from the usual venue - we should hopefully be in this new room from now on.
This will invariably be followed by drinks and chat in the pub (probably the Potting Shed).
We have two speakers:
future is an R package developed by Henrik Bengtsson, available on CRAN (https://cran.r-project.org/web/packages/future/index.html). It implements futures, variables that will be available in the future. One of the most useful way to use future is through the new assignment operator %<-%. I will demonstrate how future allows to run tasks in the background, i.e. without blocking the R terminal, as well as how future allows easy parallelisation of both homogeneous and heterogeneous tasks. future supports multithreading on Linux, OSX and Windows, as well as distributed computing.
The lasso is a powerful, widely applicable regression method that can perform both variable selection and regularization to augment the predictive and interpretive power of models. Lasso uses regularization to shrink coefficients, and is moreover able to shrink these values to zero, removing them from the model. This provides a superior selection method to stepwise approaches, which often fail to find the optimal model and can even make prediction error worse. The lasso has been applied to generalized linear models, mixed models, proportional hazard models and many others, in a variety of R packages. I will discuss a few of these, in particular, how to get started with ‘glmnet’.
For any newcomers (you’re very welcome!), here’s a map of where we’ll be.
Meetings are open for all to attend (e.g. no University of Edinburgh card required). Hope to see you there!