Skip to content

Conversation

@ctrl-z-9000-times
Copy link
Collaborator

As you can see, the scores went the wrong way.

I think part of the issue is the parameters for HTM-core are not great.

Copy link
Member

@breznak breznak left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

hmm, this is really worse. Hope we can tune htmcore's params to be competitive

@ctrl-z-9000-times
Copy link
Collaborator Author

ctrl-z-9000-times commented Sep 30, 2021

Yes, I think we definitely can do much better than these low scores...

I think some of the parameters got changed at some point; I don't think they match what numenta was using originally.

IIRC, Original Parameters are here: https://github.com/numenta/nupic/blob/master/src/nupic/frameworks/opf/common_models/anomaly_params_random_encoder/best_single_metric_anomaly_params_cpp.json

Here is the output of my new nyc_taxi.py program:

broken_tm

The TM is clearly not working. The blue line is the raw anomaly, which is almost always at 100%

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants