Hi,
I just started working on HTM for anomaly detection, I tested hotgym example and implemented it for my dataset. I created model_params using swarming. Everything is working fine, But i am not getting any anomalies detected, only first 5 values are showed as anomalies and no other value has anomaly score of more than 0.1 and 70% has 0 as anomaly score.
Also in one_gym example i am getting 0.9 as anomaly_liklihood score to most of the values.
Can anyone please help?
I just ran one_gym with this command:
python run.py
And I got this plot of anomaly likelihoods:
Does that seem like the same values you are getting?
See also What constitutes an anomaly? How to tune detection? - #16 by scott
No what i mean is, am using one_gym example on my dataset with my model_params. I have a timestamp dataset of per day values, total of 600 values. i think my model parameters are not created correctly, What to do? Please help.
If you have one value per day, then “time of day” doesn’t matter, so you can remove that encoding and replace it with a season encoding.
If there are weekly patterns, they should show up because you are still using the “day of week” and “weekend” encodings.
That being said, if you look at your data in a graph and you cannot see patterns with your own eyes, it is doubtful that NuPIC will see them either.
