Performance testing for HTM anomaly detection

I have been working with anomaly detection using HTM as part of a project. Since we need to keep a track of memory usage for the processes we are running, I was wondering if there is an estimate for the memory footprint for a piece of dataset being modelled? This can be in reference of the hotgym example (since, it has anomaly detection implemented) or any other datasets as part of the nupic.core/htm.core examples?