Most multivariate statistics and machine learning algorithms assume Euclidean metrics in unconstrained data spaces. On the other hand, most real-world data are strictly positive and capped by physical or constant-sum constraints.
Disregarding mandatory assumptions when applying statistical models may obscure meaningful patterns, produce spurious correlations, or lead to senseless measures of model quality. In this short course, useful recipes for overcoming common pitfalls in multivariate statistics and machine learning will be presented with hands-on examples.

The course is structured into three topics:
a) Why are common metrics meaningless in constrained data spaces?
b) How to deal with physically constrained data?
c) How to deal with constant sum constraints?

This course is conducted interactively, with interdisciplinary, hands-on exercises. Advanced statistical/mathematical knowledge is not required, but bringing your own laptop with an R, Python, or MATLAB environment will help you follow the recipes and exercises presented. We are looking forward to fruitful discussions and the exchange of individual challenges and experiences.