Anomalies vs predictions


#1

Can I get both anomaly detection and the HTM predicted value? Seeing what HTM would predict would help me to understand what it’s calling ‘anomalous’. Would I need to setup two networks / publishers?

Also - how does HTM know which param of the comma separated list is ‘interesting’? IE don’t waste time looking for variations in the time stamp…


#2

The MultiEncoder consists of many sub-Encoders, each with their assigned segment of the input field (i.e. the date portion could be bits 15-140 of 500). I may be wrong, but I believe that sub-group of bits is what is being “tracked” by the classifier assigned to that field. So the KEY.INFERRED_FIELDS_MAP is what is determining this. Yes you can get both Anomaly Detection and Predictions from HTM.Java’s NAPI - if that’s what you want.