I want to detect anomalies with HTM

Science Numenta Theory
1

I want to detect anomalies with HTM.
What should be set to the inferenceType of search_def.json to generate model_params.py to detect anomalies?
I tried to write Anomaly but it was not generated.
All values ​​that can be set when reading the Swarm document are separate arguments, so it is written that it is beyond the scope of this document.
I tried searching by search searching on Quick Search on the document page with search_def.json, but only the page of the same Swarm was searched

2

Have you watched this? It is still relevant, although the APIs have changed just a bit since NuPIC 1.0.

3

I am Japanese and do not speak English as my mother tongue so I can not understand even if I showed English video.
Please explain.

4

First, read the one gym prediction README. Then read the one gym anomaly README.

5

Yes I read all of README.
The prediction was completely done with.
Next, when I read the README for anomaly detection, it was written that it was just changing the inferenceType of model_paramas.py to Temporary Anomaly, so I tried it as expected, but the result was useless.
So I thought that I had to swarm by changing the inferenceType of search_def.json to Anomaly, but this time Swarm did not work well.
Where is it wrong?

6

Swarming shouldn’t actually be necessary for anomaly detection. If you use the run.py file within ‘…/opf/clients/hotgym/anomaly’ you shouldn’t need to run swarm at all because there are already parameters there. All you should have to do is:

  1. Set the ‘GYM_NAME’ equal to the name of the data file
  2. Set the ‘DATA_DIR’ equal to the directory where your data file is stored
  3. Set the ‘predictedField’ equal to the name of the variable you’re getting predictions for
  4. Make sure there is a file within the the model_params folder called GYM_NAME_model_params.py (where GYM NAME is the name of your data file
  5. Make sure that you in that model_params.py file have encoders for all variables involved.
1 Like
7

I do not understand well, but are not we going to do anomaly detection with a program listed at http://nupic.docs.numenta.org/1.0.1/quick-start/algorithms.html?

8

You can yes, though I think it is easier to download nupic (pip install nupic). Then once you have it go to this folder: ‘/nupic/examples/opf/clients/hotgym/anomaly/one_gym’. From there open the ‘run.py’ file and follow the steps I listed above. This should be easier. Also I know this is all difficult being new to English, but I think you should still watch that tutorial video @rhyolight suggested. Even if you don’t understand what he says you could still follow visually, imitate all of his commands as he types them.

9

I saw the video and did it as instructed, but nupic_anomaly_output.py can not be done anywhere.
In video it is supposed to be under your home, but it is not made.
What should I do?

10

It is in the repo right here:

11

I’m not sure why the translation of google translation is strange and I do not understand well. numenta / nupic / blob / master / examples / opf / clients / hotgym / anomaly / one_gym / nupic_anomaly_output.py
Do you mean that we need to copy?

12

According to the description of the video something like nupic_anomaly_output.py exists under my home directory, but actually it did not exist.

13

I don’t understand. The file exists exactly where it should be in the repository.

14

Yes. It is confirmed that it exists in the repository.
But from the description of the video I understood that run.py will generate nupic_anomaly_output.py in my home directory.
Is it different?
Can I copy nupic_anomaly_output.py from my repository into my home directory?

15

No, that is not true. That script is never generated. I don’t know why you would need this script in your home directory.

16

Yes. Since it explains that the description of the video is to edit nupic_anomaly_output.py under the home directory, executing run.py clarified in the sandbox under the home directory clearly generates nupic_anomaly_output.py under the home directory Because I understood that.
I understand that it will not be generated automatically.
Thank you very much.

17

I Think you misunderstand. At the begining in the video, he just copies the github directiry to home directory.

18

Sorry, it has been a long time since I created these tutorials. Did you get it working?

19

Yes. That’s right.
I overlooked what to copy.
I’m sorry.

1 Like