useR 2019 debrief

Recently I went to the useR 2019 conference in Toulouse, which is the yearly international meeting of R package developers. The talk schedule with links to slides is already online.

I gave a talk about the fast algorithm for optimal peak detection implemented in the PeakSegDisk R package, slides, video.

Here is a summary of the most interesting talks I saw during the conference:

• Torsten Hothorn, Transformation models tutorial on packages tram, mlt, trtf.
• Anqi Fu and Naras, CVXR is for disciplined convex optimization.
• Michela Battauz, video, regIRT for L1-regularized Item Response Theory models. Fused lasso type penalty that forces coefficients for all choices to go together.
• Julie Josse, video, interesting paper about supervised learning with missing values
• Eric Lecoutre, slides comparing machine learning packages.
• Erin LeDell, video, benchmarking ML systems.
• Michel Lang, video, mlr3 machine learning framework.
• Bernd Bischl, video, mlr3pipelines
• Tomas Kalibera, video, slides on sustainable package development.
• Perry de Valpine, video, slides about nCompiler, write a subset of R code that is compiled to C++ (and supports auto-differentiation). Also NIMBLE for Bayesian modeling.
• Robert Crouchley, video, slides about RcppEigenAD for auto-differentiation.
• Radford Neal, pqR now has auto-differentiation as well, blog.
• Zbynek Slajchrt, video, slides about FastR.
• Arun Srinivasan, video, slides about improvements in data.table and the .SDcols feature.
• Jim Hester, video, slides about fast vroom CSV reader.
• Gabor Csardi, video, slides about pak, parallel alternative to install.packages function.
• Henrik Bengtsson, video, slides about progressr API for progress display and future parallel computing framework.