Traning HTM


#1

good morning,
answer me please :sob: :disappointed: :pleading_face:
to detect anomaly do we train our HTM on the normal data, or we create a pattern and any data that different to this pattern considered like anomaly. and if not can anyone describe to me how the HTM detect the anomaly


#2

Yes, we do train HTM on a β€œnormal” dataset and run predictions trough it. Then see how bad the model is predicting.

This is a picture of a model predicting a normal datastream.

And when HTM is fed with a abnormal data

The red lines are the anomaly score for each time step. The blue line is the input value and the yellow dots are the predictions HTM made of what the input of the next time step.
As you can see. HTM make horrible predictions (and thus high anomaly score) when presented with abnormal inputs.

BTW, manners and proper English please


#3

thanks for your ansewr , sorry for language mistakes