Deploying R for production AND Explainable machine learning

Very welcome to our next meetup where we will address two questions that often come up when using R in a commercial setting: How to deploy R code to a production system? And how to explain to stakeholders what a prediction is based on? This time we’ll meet at IBM in Kista to learn about these topics that of course can be very relevant in other settings too.

We will start at 17.30 with mingle and refreshments (sandwiches and drinks). The first session begins at 18.00 and we will be done by 20.00. There will be a short break between the sessions.

Once again, very welcome!

Deploying R for production
Holger Hellebro, IT Architect & Data Scientist, IBM

While R can be great just running on a workstation, sometimes the real value is not unlocked until the models and visualizations are deployed as part of a system, where they can interact with other components and, for instance, process new data as it becomes available. Additionally, interactive visualizations such as Shiny apps need to be made accessible to the users to be of value. In this session we will look at some options of deploying R code to a cloud platform, using technologies such as Docker and Kubernetes. We’ll cover secure deployment of visualizations using Shiny and friends as well as making code and models available to other components as Application Programming Interfaces (APIs).

Explainable machine learning
Mikael Huss, Data Scientist, Peltarion

When building predictive models in a business context, it is often valuable to be able to give clients or other stakeholders some sort of insight into how the models make their predictions. This can be done on different levels, for instance overall variable importance measures, curves describing how the target prediction varies as a function of an input variable, or explanations of why a certain single prediction was made. I will discuss a few R packages pertinent to these questions, such as iml, pdp and lime.