nab-comportex

Support and experiments for running Comportex HTM models on the Numenta Anomaly Benchmark.

For context, read the blog post Attempting the Numenta Anomaly Benchmark.

Results

The results as submitted to the 2016 NAB Competition can be found in the results directory.

Headline results:

model	standard	low FP rate	low FN rate
original NuPIC model + anomaly likelihood	65.3	58.6	69.4
original NuPIC model, raw bursting score	52.5	41.1	58.3
selected Comportex model, delta anomaly score	64.6	58.8	69.6

Effective time step model results:

model	standard	low FP rate	low FN rate
effective time steps when 20% columns change	64.7	59.8	69.5

Usage

To reproduce the results submitted to the 2016 NAB Competition:

1. install Leiningen.
2. clone this git repo. It should be placed alongside the NAB directory so that data files can be read from and result files written to ../NAB.

• git checkout v0.1.0

3. In the NAB directory:

• python scripts/create_new_detector.py --detector comportexDepth1GlobalFrac16Stim18
• python scripts/create_new_detector.py --detector comportexDepth1GlobalFrac16Stim18NoDelta
• python scripts/create_new_detector.py --detector comportexDepth1EfftGlobalFrac16Stim18

4. In the nab-comportex directory:

• lein repl

5. In the repl: (start-notebook)
6. Note the port number such as 12345, then go to http://localhost:12345/worksheet.html?filename=worksheeets/method.clj
7. Keep pressing Shift+Enter...

The results will be written into ../NAB/results/.

It would also be possible to run the clojure file worksheets/method.clj in other ways, interactive or not.

License

Copyright © 2016 Felix Andrews

Distributed under the GNU Affero General Public Licence, Version 3, the same as Comportex.