I am very new to this package, and I read the hot gym and ny traffic examples. It seems the examples can train with the data and then detect anomalies. I have a very simple set of data as below (the whole set data is at dataset). How can I use npniu to detect the data anomalies? (In fact, if we draw 3D graph, one can see the “outliers”). Thank in advance!
Do these coordinates represent an object moving through space? Or do they represent unrelated events occurring in different places over time? There is a big difference.
You could use NuPIC and the CoordinateEncoder to encode XYZ coords over time to find anomalies in an objects path, but we so far cannot make predictions for future coordinates. If they are unrelated events, you will have less luck.
They are related events–meaning from the graph one can tell the points roughly can be grouped in 5 clusters, the first 4 of them have some uniform distributions and the last one happened in nearest time range has dispersed distribution. However, they are not moving objects ( many, many objects – a few of them might have moved).
Let me ask you this… if you had a subject matter expert on this data at hand, and they were asked to watch these events occur over time and space, would that person be able to predict future events or analyze whether an event was anomalous? This doesn’t means just spatially anomalous. To find the outliers within this data set you don’t need HTM.
What exactly are you trying to get from this data analysis?