I see a great document “the Science of anomaly detect”
In this document, the author talk about the HTM for IT as an example with many advantages compare to all other methods/algorithms . does anyone know about this app source code ( HTM for IT) ???
great. thank you @marty1885 very much, thank @rhyolight, @sheiser1 in advance, . First feel, all of you and orther members in this forum are so kind, ready to help and very friendly. I hope i will be able to support you all soon if possible.
You are in the right place at https://github.com/numenta/numenta-apps. Look for the word “taurus”, which is the codeword for the HTM for Analytics demos. In this example, we’re analyzing stock volume instead of server stats.
This code is old and no longer maintained. It is also Python 2.7. We are not planning to continue working with it.
Study and evaluate, kinda. The repo is in Python 2 and it will be EOLed in 3.5 months.
However you should not deploy the repository. You’ll either need to migrate the repo to Python 3 or built your own in Python 3.
Yes, so please use HTM.core, the community fork of NuPIC.
Or use Etaler, my HTM implementation aimed for HPC applications (It’s C++ only for now, Python wrapper will come later).
But after them SP, TM they use the SDR Prediction Classifier like this
Get the bucket info for this input value for classification.
bucketIdx = scalarEncoder.getBucketIndices(consumption)[0]
# Run classifier to translate active cells back to scalar value.
classifierResult = classifier.compute(
recordNum=count,
patternNZ=activeCells,
classification={
"bucketIdx": bucketIdx,
"actValue": consumption
},
learn=True,
infer=True
)
# Print the best prediction for 1 step out.
oneStepConfidence, oneStep = sorted(
zip(classifierResult[1], classifierResult["actualValues"]),
reverse=True
)[0]
print("1-step: {:16} ({:4.4}%)".format(oneStep, oneStepConfidence * 100))
results.append([oneStep, oneStepConfidence * 100, None, None])
return results
It make me confuse because:
My First Goal is “anomaly detection” .
SDR Classifier is new for me.
In your advice, if my goal is anomaly detection then do i need to do the SDR Predictionn Classifier or not ? Or bypass thisn( SDR Classifier) and go to anomaly detection branch ? Classifier Prediction and Anomaly detection are different branches ?
As I understand the Classifier is only needed for forecasting, and can be skipped if your sole purpose is anomaly detection. From the TM you get the raw anomaly scores, which you can then use to get the anomaly likelihoods.