Hi @immortalluo, welcome to the forums! I’m not the most experienced here, but I’ll have a shot at helping you - hopefully I don’t lead you astray
The particular parameters you are looking at relate to what constitutes an anomaly in a stream of data. HTM makes predictions based on sequences, but whether or not an incorrect prediction should be interpreted as an anomaly is open to some tuning.
I would suggest a quick read of this paper, which sets the scene for applying HTM to the problem of anomaly detection: https://numenta.com/assets/pdf/whitepapers/Numenta%20White%20Paper%20-%20Science%20of%20Anomaly%20Detection.pdf
In particular, page 3 shows a box: “Prediction Anomaly Detection Classification” which takes the output from the HTM learning algorithms.
You can see this relationship nicely in HTM Studio - there is a stream of data that HTM is learning to predict the next value of, then separately there’s a classifier which determines the level of anomaly based on the success of the prediction:
I’ve never tweaked the parameters you’ve mentioned so I don’t know offhand exactly what they do, but there are some comments in the code that might help: https://github.com/numenta/nupic/blob/master/src/nupic/frameworks/opf/htm_prediction_model_classifier_helper.py