What constitutes an anomaly? How to tune detection?

scott · May 22, 2017, 9:56pm

We typically threshold the likelihood with four (possibly anomaly) or five (definitely anomaly) 9’s. We also have an additional step for computing the log likelihood, which converts 0.99999 to 0.5 and 0.9999 to 0.4. The log likelihood is suitable for plotting, while the straight likelihood is not (as you demonstrated). See here for converting the log likelihood:

github.com

numenta/nupic/blob/99856ea00fc76f9a3fbb86ad9703649750cb6d5d/src/nupic/algorithms/anomaly_likelihood.py#L220-L230


@staticmethod
def computeLogLikelihood(likelihood):
  """
  Compute a log scale representation of the likelihood value. Since the
  likelihood computations return low probabilities that often go into four 9's
  or five 9's, a log value is more useful for visualization, thresholding,
  etc.
  """
  # The log formula is:
  #     Math.log(1.0000000001 - likelihood) / Math.log(1.0 - 0.9999999999)
  return math.log(1.0000000001 - likelihood) / -23.02585084720009

Topic		Replies	Views
Is HTM right tool for my case? (Online anomaly value detection in data stream) HTM.Java	4	697	September 26, 2018
Finding the predictability of a pattern of anomalies with HTM NuPIC	4	462	April 10, 2019
Anomaly score/likelihood question NuPIC Community Fork question	6	735	September 2, 2020
Inability to accumulate minor anomalies? NuPIC encoders , anomaly-detection	9	1448	November 10, 2016
HTM detecred an anomaly. How to find out the reason? Lounge	3	596	August 30, 2017

What constitutes an anomaly? How to tune detection?

Related topics