Nupic Anomoly Detection


Over the weekend I got the python 2 version of Nupic ( up and running and ran the Hot Gym Prediction / Anomaly which completed successfully.

When I ran some new data through the prediction (after swarming and selecting a model), using the same columns of date and a floating point value, the prediction values exactly matched the actual values for 80% of the 10,000 rows. I suspect that this means that there was actually no prediction for these rows - is this a correct suspicion?


Basically correct. If there is not enough certainty to create a prediction, I think the OPF simply returns the last input data value.

1 Like